Home

Java Magazine - March/April 2015

1. MARCH APRIL 2015 kJ u p a a IF Netflix powers through 2 billion content E per day with Java driven architecture A eo mm c PP a E Y 3 me e S f E A ta pa dm id zw pa i om P dv aas le ORACLE COM JAVAMAGAZINE ORACLE table of contents NETFLIX 21 ACTIONI Netflix moves fast to provide the viewing experiences its customers want Java helps make it happen Java and Performance New theme icon See how it works COVER ART BY PHIL SALTONSTALL COMMUNITY JAVA IN ACTION 03 12 From the Editor loT Developer Challenge 05 Winner Java Nation Control your greenhouse environment automatically with Smart Greenhouse News people books and events 14 JAVA TECH N JCP Executive Series 33 Performance Innovation N Java Architect and Success Understanding Java JIT Intel s Anil Kumar describes how Compilation with JITWatch collaboration drives a healthier See the effects of small source code Java ecosystem changes and Java HotSpot VM switches ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 a 2 m Java Architect IMPROVING THE OBSERVABILIT Y OF FORK JOIN OPERATIONS How to determine when adding parallelism might help performance 40 Java Architect The Quantum Physics of Java An introduction to modern chip design and its effect on Java programs 47 Java Architect Shakespeare Plays Scrabble New ways of solving problems with the
2. BENCHMARKMODE SPECIFIES WHAT SHOULD BE MEASURED FOR EXAMPLE THROUGHPUT OR AVERAGE TIME ee c009092029206060629 62060602062062 62062920620606296292062 62 62o0pt 09292 6 272722 02 0229220 0292 0206292 6020629296929620620 62 629 62090206020620620 6020620620602952292060292292090290296206020206020602092902920602929606092962060 20 629209020 60296206006029 229290260292902902602092929602062929620 620692962062092920620602060296290602962062929 229 06029229229292960206029029202 62 292960 09 0299 9299999 OUTPUTTIMEUNIT SPECIFIES THE TIME UNIT USED IN THE OUTPUT FOR EXAMPLE TimeUnit MILLISECONDS eec oo oo o o e e o e e ne foscoc Sc o ooo eee eee reece eee eee eee e eno e e e eee e e ne e e e e e ee e e eeo ee oc o o o np o e e e e o AWARMUP SPECIFIES THE WARMUP PHASE YOU CAN SET THE NUMBER OF ITERATIONS AND THE DURATION OF A SINGLE ITERATION 3 Billion MEASUREMENT SPECIFIES THE MEASUREMENT PHASE AND IS SIMILAR TO WARMUP Devices Ru n Java SPECIFIES HOW OFTEN YOU WANT TO FORK THE JAVA VIRTUAL MACHINE JVM AND RUN THE AFORK TESTS YOU SHOULD ALWAYS DO RUNS IN SEVERAL FORKS Table 4 Tests written for JMH are simi focused on memory and spe lar to JUnit tests The code you cifically the cache hierarchy The want to benchmark needs to be in second part looks a little more into a single method which must be
3. 33 iload i20 Has ad euismod evertitur pertinacia at labore omittam duo 135 bipush 21 Eam dicat eligendi pericula in Vel cibo probatus in 37 at icmpgt 22 Mei ea impetus persius et facer laudem minimum mel 40 iload 23 l42 bipush 24 Benchmark 144 13ub 25 public String convertString Blackhole bh 145 12e 26 return SOURCE toUpperCase l 446 istore 127 148 aload_3 2B 149 iload 29 dHenchmark i51 iload 130 publie String convertCustom Blackhole bh 1531 castore 131 return toUpperCaseAStii SOURCE 54 sine 32 is goto 33 60 new 34 public final String toUppertaseASCII String source 163 dup 35 int lan source lengthi i64 aload 3 36 65 invokespecial 37 char result new char len 168 areturn EE 139 for int i Q i len i 140 Figure 7 Compile time ms 6 Method java lang String length I di ES 12 8 class java lang String 89 Method java lang String init C V ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 rather than from the Sandbox and load the resulting Java HotSpot VM log file into JITWatch for analysis The code for this benchmark is shown in Listings 2a and 2b It turns out there is a heavy price to pay for supporting all character sets The custom version achieved more than 4 5 times as many oper ations per second as the core library version as shown in Listing 3 Analyzing the log file from JMH shows that the me
4. developed a vacuum cleaner bag finder on the internet You entered your vac uum brand and model and it spit out the fitting bag type from the prod uct line Java Magazine What do you enjoy for fun and relaxation Eisele I m most relaxed and happy in front of a computer But l enjoy social interaction as well Java Magazine What happens on your typical day off work Eisele Since I love my work there s no real day off I catch up with news from the Java community and my JBoss peers daily If the weather is fine we try to make something exciting happen for the kids theme parks or stuff like that Java Magazine What side effects of your career do you enjoy the most Eisele Traveling I love experiencing countries people habits and culture talking about technology and engag ing with people Twitter and blogs are fine but traveling helps me understand the culture of countries soI can work effectively with developers and custom ers there Java Magazine Has being a Java Champion changed anything for you with respect to your daily life Eisele Yes mostly from Oracle s handling of the Oracle ACE and Java Champion community programs They sup port us by maintaining a strong brand and help ing us make the right connections Java Magazine What are you looking forward to in the coming years Eisele The second edi tion of JavaLand which I want to be even big ger and better t
5. new Quartar 17 1H Coin coin null 19 20 change the variable maxImplementations to control the inlining behaviour 21 2 himorphic dispatch the method call will be inlined 2 3 megamorphic dispatch the method call will not be inlined 23 24 Final int maxImplementations 2 25 26 For int i 0 i lt 100000 i 27 switch i maximplaementations 2B case coin nickel break 29 case l coin dime break 30 case 2 coin quarter break 31 32 33 coin depositi 34 35 J6 System out println moneyBox moneyBox 37 38 30 public static void main String arga 20 new PolymorphismTest iz Figure gure 3 See the listing shown in Figure 3 On line 24 you can control the number of implementations that are used by setting the variable maxImplementations Click the Run button in the Sandbox window to execute the program The JITWatch TriView window shown in Figure 4 opens as soon as the Java HotSpot VM log file has been analyzed When two of the Coin implemen tations are used hovering over the invokeinterface instruction at offset 93 in the Bytecode pane shows a tool tip confirming that the method java architect v Source Bytecode v Assembly Compile Chain JIT Journal Bytecode size Native size Compile time ms AAA ES Source yte E n n 5 public static int moneyBox 0 15 istore 5 6 3 37 iconst 0 public
6. Hive used to have to write a lot of intermediate data out to disk between each MapReduce step Tez can remove a lot of this overhead if you turn it on Long running Tez sessions and reuse of containers can take several seconds off the startup time and a wide range of tuning options also help reduce overhead from the Java Virtual Machine JVM startup Most distributions don t turn on all the optimizations so the most common cause of overhead at the moment is just not turning on the new features Question Can you talk about the relationship between the Lambda Architecture and Hadoop Elliston Ball The Lambda Architecture is a much debated topic in the big data community at the moment Even before the ink dried on Nathan Marz and James Warren s book Big Data Principles and best practices of scalable realtime data systems which outlined the architecture developers were lining up on either side to critique it The approach works by separating the batch and speed layer of your applica tion Hadoop has traditionally been all about batch processing so for abso lutely up to the second analytics you needed a streaming system on top to do intermediate updates on the earlier batch results A serving layer would then combine them The best use for the Lambda Architecture is when you have a slow running algorithm running in a batch across a large data set which can be approximated by a short running ver sion over a stream
7. of Java processes running all the time yet as we grow we don t have huge infrastructure challenges Glover says f T TL EL ATTI HI iid j ij IPR TIER I ff li I I He Hi HH for the continuous delivery and deploy ment of new software within the vast worldwide Netflix infrastructure Rapid turnaround is an important differentia tor for IT and itis certainly a competitive advantage for Netflix he emphasizes Thanks to our continuous delivery platform if a business unit wants a new feature pushed into production we can do it within a couple hours Glover is echoing one of the core mantras at Netflix move fast make decisions fast and do it all on a solid JAZINE MMU MARCH APRIL 2015 architecture Because we are a data company we need to be able to facili tate rapid innovation across the globe he adds We need to make decisions fast based on the data we collect so we can provide the viewing experiences that our customers want and stay ahead of competitors HOW NETFLIX WORKS Netflix members love being able to browse select and play movies with a few clicks of a remote control or mouse ATING INNOVATION ake decisions fast based on the data we collect Andy Glover F A es L T z e need tom but not many of them stop to think about the sophisticated hardware and software infrastructure that delivers the content Whenever you launch Netflix to queue up Ca
8. uim Click for tips on viewing the stereogram e Atalaso Image enabling experts amp bacon connoisseurs Visit us online to see our full line of SDK products for NET and Java O jt www atalasoft com java nation er ma IW E E rai wee i I LE Ru M reir it mi gt f ar e jiri E aw s iE f E i i m E Pe rea mu JCER TIF TRAINING DEVELOPERS IN AFRICA PHOTOGRAPHS COURTESY OF JCERTIF INTERNATIONAL ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 The JCertif Kigali Java developer conference was a huge success sharpening the skills of hundreds of Java programmers JCertif Inter national has plans to expand its events to more African nations gt E O O d A T f d lira id s i ngs pea n Vie p Oe E A winner of the 2014 Duke s Choice Awards Certif International has trained more than 5 000 new Java developers across Africa since its start in 2010 JCertif Kigali and JCertif Tunisia were just its latest great successes JCertif Kigali was the leading devel oper community conference in Rwanda Founder and Managing Director Max Bonbhel says We re still digesting the great num ber of Likes collected during the event The events are two to five day Java workshops for groups of 20 to 50 developers JCertif is planning a series of events in 2015 including in Cameroon Benin Senegal the two Congos Ivory
9. 21 2 A 6 65 81 9 4 FORK JOIN THREADS 20 14 0 3 60 3 88 12 0 Table 3 of threads by setting the property Djava util concurrentFork oinPool common parallelism N Table 3 shows the results of the mixed workload with 4 8 and 16 instances of ForkJoinWorkerThreads The results in Table 3 confirmed my suspicion that adding more threads would not have any sig nificant benefit The CPU utiliza tion chart in Figure 1 offers some explanation of why this might be the case The two tasks were submit ted such that subtasks from both workloads should have intermixed However there is a telling sign in Figure 1 the big red blobs due to overloading the kernel with disk I O operations The first run with 16 threads is domi nated by kernel activity In contrast the subse quent runs with 8 and 4 worker threads show much less kernel activity able to understand how e Censum gt I should remind you that the fork join framework e ForkJoinPoolMonitorMXBean the exact same work performed by adding project on GitHub ORACLE 32 ADDING THREADS load was offered up in all three cases The only difference is the number of threads in the common thread pool Conclusion What these experiments indi cate is that there doesn t seem to be much benefit in adding more threads In fact these results show that adding more threads and hence adding more pressure on an already stressed I O system can be exceptionally detrimental to
10. 8 16 32 64 128256512 1 2 4 8 16 32 64 128256512 KB KB KB KB KB KB KB KB KB KB MB MB MB MB MB MB MB MB MB MB Figure 5 that is more accessible to human CACHE beings my former colleague Richard Thompson came up with HIERARCHY MN uuum Figure 6 times Ll cache is the smallest and fastest Current CPUs typi cally have three cache levels with each level being slower and signifi cantly larger than the level before Figure 6 shows an improved ver sion of our model that contains the cache hierarchy How large are the performance gaps between the different cache levels To explain this in a form the beer cache hierarchy Imagine that you re sitting in front of your TV watching your favorite team and you re thirsty Ll cache is the bottle of beer in your hand Access time is almost immediate 1 ns but the quantity is extremely limited for example 32 KB on my system L2 cache is the cooler next to your sofa Access time is still pretty low 7 ns and the quan tity is significantly larger 256 KB which is equivalent to 8 bottles of beer L3 cache is the fridge in the kitchen Access time is noticeably ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 public int rndIndex Setup Level Iteration public void setUp rndIndex new int indexes A final List lt Integer gt list new ArrayList lt gt indexes for int i 0 i lt indexes i list add 16 i Collections
11. Android appli cation has more files and directo ries which allows for more custom ization of the packages However in the HelloWorld demo everything is kept to a minimum One of the benefits of the Gradle system is that it relies on the convention over configuration principle The default behavior is sufficient in most cases but there are tons of configuration options so you can configure the application the way you want it The AndroidManifest xml file in the src main directory can be used to supply information about the application At a mini mum the information shown in Listing 2 must be specified in this manifest file The text in italics should be altered for each application Note that the android name part of the activity tag can be chosen freely MAGAZINE FMIMIMIMIIMIMMB MB MEMIIMMMMMMIMIMITIT MARCH APRIL 2015 apply plugin application mainClassName org javafxports helloworld HelloAndroid dependencies compile project core 3 Download all listings in this issue as text and it is not necessarily the same as the main class attribute specified in the meta data section of the activity tag The latter has to be the name of the JavaFX application class In the build gradle file we have to instruct the Android tools to take into account the JavaFX runtime libraries both the Java classes and the native libraries The build gradle file contains a dependency section in which we have to declar
12. Eastern US and EMEA APAC Sprocket Media Mark Makinney 1 805 709 4745 Advertising Sales Assistant Cindy Elhaj 1 626 396 9400 x 201 Mailing List Rentals Contact your sales representative RESOURCES Oracle Products 1 800 367 8674 US Canada Oracle Services 1 888 283 0991 US Oracle Press Books oraclepressbooks com If you are interested in submitting an article please e mail the editors SUBSCRIPTION INFORMATION Subscriptions are complimentary for qualified individuals who complete the subscription form MAGAZINE CUSTOMER SERVICE java halldata ccom Phone 1 847 63 9635 PRIVACY Oracle Publishing allows sharing of its mailing list with selected third parties If you prefer that your mailing address or e mail address not be included in this program contact Customer Service Copyright 2015 Oracle and or its affiliates All Rights Reserved No part of this publication may be reprinted or otherwise reproduced without permission from the editors JAVA MAGAZINE IS PROVIDED ON AN AS IS BASIS ORACLE EXPRESSLY DISCLAIMS ALL WARRANTIES WHETHER EXPRESS OR IMPLIED IN NO EVENT SHALL ORACLE BE LIABLE FOR ANY DAMAGES OF ANY KIND ARISING FROM YOUR USE OF OR RELIANCE ON ANY INFORMATION PROVIDED HEREIN The information is intended to outline our general product direction It is intended for information purposes only and may not be incorporated into any contract It is not a commitment to deliver any material code or func
13. Generations B Relative Time Spent in G 10 05 AM 1C carefully ensuring that proper resource handling is used An application s performance tuning should begin with the source code Even though the top causes of performance issues in Java EE applications point to the environment proper coding can still play a key role in an application that performs well The following poor coding prac tices among others can lead to performance issues Over serialization and deserialization Overuse of finalizers Too much synchronization Not discarding unused variables Too many dynamic variables Rampant use of System out printIn Sessions that are not released when they are no longer needed Failing to close resources for example database and network connections Performing code reviews is imperative for reducing code that inhibits an application s perfor mance While poorly crafted code can slip past multiple develop ers the more eyes that examine it the better In addition to code reviews more than one individual should run performance and load tests against an application and compare results of current tests to those run previously JAVA IN ACTION am O LLJ lt lt ABOUT US enterprise java M ES org javaee7 concurrency ES org javaee7 entity E org javaee interceptor ES org javaee7 interfaces ES org javaee7 jms L4 E org javaee7 jsf HJ org javaee7 listene
14. Israeli software landscape PHOTOGRAPHS BY ILAN SHACHAM GETTY IMAGES AND TOMER GABEL FEATURED JAVA USER GROUP which includes young startups established enterprises and consulting firms Gabel says Between 50 and 100 members participate in monthly meetups Gabel describes Java IL meetups as focusing largely on production war stories whether it s an alternative language for example Clojure a brand new in memory database or an amaz ing REST documentation framework we seek and promote speakers with hands on produc tion experience Sharing the real caveats can save an immense amount of time for partici pants who are often afraid to take the plunge without knowing the risks The Java IL leadership takes a genuine interest in career development among the community including developing presentation skills While ORACLE COM JAVAMAGAZINE MIMMMIMMMM MP BM M PM MARCH APRIL 2015 gt Z gt 2 2 O O we strive to get as many people as possible from the Israeli community on stage to showcase their projects not all of us are accomplished public speakers says Gabel We therefore offer guidance and assistance in everything from initial preparation and abstract through developing a slide deck to providing an oppor tunity for dry runs of the lecture Although Java IL is relatively new the group is already involved in regional events For example Java IL organized the back end track at CodeMotion
15. Java Run the command shown in Listing 1 This command might take a couple of minutes to complete because the tool disassembles each method in the JAR file s class files using the API form of the javap program called com sun tools javap JavapTask In Java SE Development Kit 8 Update 31 JDK 8u31 on 64 bit Linux 3 605 methods exceed the FregInlineSize threshold and many of these aren t a sur prise For example java util GregorianCalendar computeFields has a lot of work to do that justifies its 1 571 bytes of bytecode Some of the results however might raise an eyebrow because they include core library methods such as java lang String toUpperCase and java lang String toLowerCase that you might expect to find inside tight loops These both weigh in at 439 bytes of bytecode The Javadoc explains that this is due to the requirement to sup port character sets in all lan guages Since case mappings are not always 1 1 char mappings the resulting String may be a different ZMU MARCH APRIL 2015 STINGI LISTING 22 USTING 2b LISTING 3 JjarScan sh home chris jdk1 8 0_31 jre lib rt jar gt methods csv nd Download all listings in this issue as text java architect length than the original So they contain code to check and resize the underlying character arrays if needed This spawned the idea that if your application domain doesn t require the flexibility the core libraries offer and per
16. Stream API in Java SE 8 57 Enterprise Java Improving the Performance of Java EE Applications Incorporate performance tuning into your development lifecycle 9l Enterprise Java BIG DATA AND JAVA Our panel of experts answers pressing questions about working with Java and big data 65 Rich Client Writing JavaFX Applications for Mobile Devices With recent JavaFX ports you can now create Java client applications for mobile devices 69 Fix This Take our Date and Time API code challenge EDITORIAL Interim Editor in Chief Kay Keppler Community Editor Yolande Poirier Java in Action Editor Michelle Kovac Technology Editor Tori Wieldt Contributing Writer Kevin Farnham Contributing Editors Claire Breen Blair Campbell Stephen Chin Kay Keppler Karen Perkins Reza Rahman Simon Ritter James Weaver DESIGN Senior Creative Director Francisco G Delgadillo Design Director Richard Merch n Contributing Designers Jaime Ferrand Diane Murray Arianna Pucherelli Production Designers Sheila Brennan Kathy Cygnarowicz ARTICLE SUBMISSION PUBLISHING Publisher Jennifer Hamilton 1 650 506 3794 Associate Publisher and Audience Development Director Karin Kinnear 1 650 506 1985 Audience Development Manager Jennifer Kurtz ADVERTISING SALES President Sprocket Media Kyle Walkenhorst 1 323 340 8585 Western and Central US LAD and Canada Sprocket Media Tom Cometa 1 510 339 2403
17. a JavaFX appli cation In some cases you want to do the opposite You have an Android application and you want to use JavaFX components inside your Android application While this adds more requirements for the integration of user interfaces itis technically possible The samples repository available here contains the Kokos project which is an example of doing this This example was created with the Android Studio IDE and it starts MAGAZINE FMIMIMIMIIMIMMB MB MEMIIMMMMMMIMIMITIT MARCH APRIL 2015 P LISTINGS LISTINGS android content Context ctx FXActivity getInstance android nfc NfcAdapter nfcAdapter NfcAdapter getDefaultAdapter ctx nd Download all listings in this issue as text with a regular Android applica tion By using the Android Studio interface you created a new code fragment code as a sub class of javafxports android FXFragment The FXFragment class is a sub class of the Android Fragment class and it allows you to run a JavaFX application inside its view In the Kokos example the javafxports org kokos MyFxApp class is a JavaFX application that is assigned to the created fragment as shown in Listing 6 Note that the possibility of using JavaFX application code inside Anaroid fragments is still under early development Start with a JavaFX application whenever pos sible rather than starting with an Android application and adding a JavaFX application to it Conclusion Jav
18. causes of a stalled front end cycle is a cache miss which results in the CPU waiting for data to arrive from main memory or a slower cache To validate this assumption you can run perf again but this time to specifically measure cache loads and cache misses You can see the result in Table 3 The ratio between success ful cache loads and cache misses differs tremendously When accessing the array in sequential order only about 6 percent of all memory loads result in a cache miss But when accessing the 19 429 967 708 311496 19 006 043 778 97 82 4 34 18 296 349 545 94 17 170 633 221 6 27 369 999 699 214 44 JRACLE COM JAVAMAGAZINE YUMI MARCH APRIL 2015 array in random order two out of three loads result in a cache miss The high number of cache misses is expected because the array doesn t fit into the cache and you have to load everything from main memory But why are there almost no cache misses when you access the array sequentially Loading data from main memory into the cache is often a major bottleneck For this reason the CPU tries to help by guessing which data you ll use next and loading it into the cache in the background as you can see in Figure 9 While you modify the elements of a cache line functionality called the prefetcher loads the next cache line into the cache Thus when you need the data it s already available in the cache The prefetcher is not par ticularly sm
19. class Nickel implements Coin public void deposit moneyBox 5 JB istore E 0x0000000102496682 mov Arlid 0x9B B 50 iloa E 9 public class Dime implements Coin public void deposit moneyBox 10 42 ldc 28 ff int 1i 1 44 if icmpge 104 0 x000000010249c689 jmpq 020000000 111 public class Quarter implements Coin public void deposit moneyBox 25 47 iloa 6 0x000000010249c68e cmp rsi r10 12 4 amp 9 iconst 2 UxdD00D00D10490691 ne Oxd0DO000010 113 public PolymorphismTest 50 irem 14 Coin nickel new Nickell Sl tableswitch fi U to 2 c693 115 Coin dime new Dime 0 76 000000001 02490697 mov irlld 0x9B 16 Coin quarter new Quarter 1 82 3 I7 2 BH 18 Coin coin null default 91 0x000000010249c6 e jmp 0x000000010 ho 0x0000000102 8c6a0 mov SOxELEELLES 120 change the variable maxImplementations to control the inlining behaviour 16 aload 1 0x000000010249c6a5 xchg fax tax i21 2 bimorphic dispatch the method call will be inlined TT astore 4 0x000000010249c6a7 callq 0x0000000 22 3 megamorphic dispatch the method call will not be inlined 19 gato a1 23 B2 aload 2 194 Final int maximplementations 2 B1 astore 125 H5 gota 91 0x000000010249c6ac calig 0x0000000 26 For int i 0 i 100000 i BB aload 3 0x000000010249c6bl mov r123d ri0d 2 7 switch i maxImplementations B9 astore 0x0D00000010249c55b4 mov Srifd triad 28 case 0 coin nickel break 91 aload 4
20. commodity hardware running Linux Ubuntu the future It also makes it easy to hire skilled engineers When it comes to recruiting it is easy to find Java devel opers Glover notes We can always find top talent because we are not chasing some brand new technology or language that only hipsters know Glover believes Java 8 has revolution ized Java calling it the most significant release of Java in the last 1O years He especially likes default methods lamb n i dim ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 das and the java util stream API which enables functional style operations on streams of elements such as Hadoop MapReduce transformations on collections Functional style program ming as a core language feature is a powerful construct that will have a big impact on the developmental landscape he predicts Java 8 dra matically changes the Java language for the better A EX lt _ o HR Netflix uses Spring Boot as the basis of its SOA because it offers the scalabil ity and maturity of the JVM Netflix is a giant SOA Glover says Java platform services make it possible for a devel oper to quickly come up to speed and write a service that works in our archi tecture Our operational footprint on the JVM is tried and true Spring Boot is a good example of a web framework that you can get up and running in short order There is still a tremendous amount of innovation w
21. culprit for application performance degradation In the case of GlassFish the server contains built in monitor ing APIs so you can monitor the modules of your Java EE application server Monitoring can be config ured with the asadmin command MARCH APRIL 2015 line utility or through the adminis trative console shown in Figure 9 Each of the modules can be configured to OFF the default LOW or HIGH After the modules are configured an application can be monitored at the console see Figure 10 or with the asadmin util ity The asadmin utility provides the added benefit of letting you pipe output into a file or a database to be analyzed at a later date against other saved monitoring data Applications and server mod ules can also be monitored on the web at the default URL of http localhost 4848 monitoring domain server as seen in Figure 11 Because the GlassFish monitoring API is REST based applications can be built to interrogate the data and display in useful for mats One such utility is LightFish which Java EE expert Adam Bien wrote If you use another Java EE application server search for the monitoring tool that best suits the environment Careful monitoring of the appli cation server using monitoring resources such as those offered by GlassFish is key to pinpoint ing performance issues in Java EE applications after deployment You must also monitor server resources such as memory and CPU
22. deposit can t be inlined ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 java architect inline success reason inline hot predicted call klasss 7Bl1 Bei 94 gt 3 3 lt predicted call klass 782 bci 93 gt lt parse gt parse method 792 lt parse gt lt bc code 178 bcie 104 method bytes 9 names deposit flags 1 holder 782 lt call method 793 inline 1 count 11264 prof factor 1 gt id 793 lt uncommon_ trap reason null check bci 93 action maybe recompile gt O uncommon trap reason bimorphie bei 93 action maybe recompile gt lt parse method 793 stampe 0 704 uses 11264 lt void PolymorphismTest Nig parse done nodes 264 memory 66184 stamp 0 704 live 258 stamp 0 704 uses 11264 lt void PolymorphismTest Dim parse done nodes 261 memory 60680 stamp 0 704 liwe 274 gt P TT e ee 2 25 1222 1 Se See Nee 88 aada dA jsi EJAAN cede ee Bee A ear M 1 ml ae iicount 282 and there is no inlining tool tip at bytecode offset 93 as shown in Figure 6 In addition you won t find the inlined assembly code for deposit as was seen in Figure 4 and the virtual call for the invokeinterface bytecode hasn t been eliminated Also note that the native code size of 504 bytes is smaller than with Henn i bimorphic dispatch because no inlining has occurred to increase the size of the native code T
23. each 4 bytes long The limiting factor in this setup is loading the cache line into the cache it almost doesn t matter how many operations you execute on each cache line This explains why the performance of both loops is roughly the same Counting instructions to estimate the performance of an algorithm is java architect Figure 4 a useful approximation because it s easy to assess and usually gives a good indication But as this exam ple shows you have to keep in mind that it s just an approximation In reality the execution times of single instructions vary widely and you can t rely on this number only Data Size Does the size of a data structure affect performance To answer this question run a small experiment using the code in Listing 2 Take the second loop from the first code example and run it repetitively This time change the size of the array and measure the average time to run a single loop iteration The purpose of this experiment is to run a trivial algorithm over a data structure whose size you can con trol Is there a relationship between the size of the array and the time needed to modify a single element Before you look at the results consider briefly what you expect Accessing a single array element requires constant time O 1 Thus the inner part of the loop should be executed in constant time too That means for large enough arrays you will hit an upper bound that is const
24. f elit O A A A A workloads under these different CONCURRENT PARALLEL 20 3 2 1 9 1 39 conditions can be found in Table 1 fete tater tetas EM cc coc c qu CONCURRENT FLOOD 20 6 0 1 9 3 2 38 which shows GC log parsing exclud FORK JOIN ing I O operations and Table 2 1 A ce eee eaeeeee ee eee eee which shows GC log parsing includ CONCURRENT FLOOD 0 2 0 0 4 ing I O operations STREAM Before we look at the results Table 1 further I should give a word of caution There is enough noise in these experiments that I wouldn t put any stock in the fine differ ences The variation between runs LAMBDA PARALLEL 20 2 8 2 8 JD was on the order of several sec LAMBDA SERIAL 0 1 9 0 0 130 onds which is why the results are SEQUENTIAL PARALLEL 20 26 26 52 coarsened to two significa nt digits mp Hl That said the benchmark appears CONCURRENT PARALLEL 20 5 8 30 19 60 to be good enough to answer the CONCURRENTFLOOD 20 43 6 5 6 6 130 following question Can we do FORK JOIN better by adapting the size of the e N S A M r E MP e mr Ce T aM e c eR eH ForkJoinPool The answer in this CONCURRENT FLOOD 0 3 0 0 i case is mixed STREAM Parallelizing brings benefits Table 2 MI PM MB MM MM MB MARCH APRIL 2015 java architect CREATE IHE FUIURE oracle com Java 16 FORK JOIN THREADS 20 210 6 13 2 13 83 304 9 8 FORK JOIN THREADS 20
25. have large ecosystems As a result we have multiple coengineering efforts with teams working at varying levels Some are focused at the compo nent level while others operate at the strategy level But the core goal is to maximize performance and to seam lessly enable these ecosystems From Intel s perspective performance means the complete end user experi ence response time certainly but also ease of use and interoperability between the components within these environments The computationally intensive application space is mature but still growing This includes every thing from weather prediction to graph simulations chemical and atomic reactions modeling life sciences and statistical process control The cloud plays an important tech nology role enormous numbers of mobile devices require processing and storage in the cloud So Intel and Oracle also collaborate there to better ensure security reliability and performance And the IoT has seen explosive growth in the past few years Many players both hardware and software occupy this space But there is also a great deal of fragmentation with differ ent approaches different types of soft ware and different hardware environ ments It will be important for Intel and ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 Oracle to collaborate around the evo lution of standards in this space and to develop effective strategies around such collaborations But
26. of updates Periodic runs of the batch approach are then used to get a more precise answer An example is a recommender system that performs complex math over your entire buying history which can be incrementally improved using a simpler formula on today s transactions and then be brought in with the batch algo rithm by an overnight job Hadoop provides a perfect plat form for implementing the Lambda Architecture The batch layer can be done in MapReduce or something like Hive and Pig while Apache Storm provides a good platform to imple ment the real time speed layer using HBase as the serving layer With YARN all this lives on one cluster Spark and Spark Streaming also provides a great framework for the Lambda Architecture and because of its micro batch approach it might allow some reuse of code between the layers However if you are reusing a lot of code you re probably not using the Lambda Architecture as it was intended Question Is there Hadoop as a Service Babadopulos Yes some companies pro vide Hadoop as a service for example Amazon EMR deploys a Hadoop cluster inside an Amazon EC2 infrastructure S E NARDON Many companies use Hadoop in production right now to run important services PHOTOGRAPH BY BOB ADLER GETTY IMAGES Qubole can deploy Hadoop as a ser vice and many other tools around the Hadoop ecosystem such as Hive Pig Apache Oozie and Sqoop either usi
27. publications as Wired Rolling Stone Playboy SF Weekly and the San Francisco Examiner He recently published a science adventure novel The Shroud and regularly contrib utes to The Huffington Post LEARN MORE e The Java Community Process 10 Your Destination for Java Expertise Written by leading Java experts Oracle Press books offer the most definitive complete and up to date coverage of Java avallable Programmer I amp Il Study Guide Exams 120 803 amp 120 804 Complete Exam Preparation Kathy Sierra Soir Bert Bates se oca ccr ku Kathy Sierra Bert Bates This self study tool offers full coverage of OCA OCP Exams 1Z0 803 and 120 804 Electronic content includes 500 practice exam questions Enterprise Architect Exam Guide gt lava Exams 120 807 120 885 and 170 866 Complete Exam Preparation Pau R Allen lr Joseph J Bambara Paul R Allen Joseph J Bambara Get complete details on Oracle Certified Master Java EE 6 Enterprise Architect Exams 1Z0 807 1Z0 865 and 1Z0 866 120 multiple choice practice exam questions are included Mastering Lambdas Applications ava Programming ina i on java Multicore World Develop Deploy Monitor and Manage Your wll d IC P Lees NN ava Cloud Applications and Streams ur eres e voa Mns B aca Dus uve C Harshad Oak rents ACE krepi imd Sea Chetan Harshad Oak Build highly available scalable secure distributed applications on
28. the good news is that Intel and Oracle have a long history of fruitful collaborations If we look at standard benchmark data for example specifi cation JBB Java Business Benchmark we see that hardware and software total performance increased 30 times from 2007 to 2013 And such collabora tions have benefits far beyond simple performance in terms of the evolu tion of Java standards When we work together around standards it creates a healthier ecosystem more competi tion which drives innovation and which Left to right Eric Kaczmarek Pranita Maldikar Anil Kumar and Soji Denloye collaborate informally gt 2 2 O O 2 n p Po gerer UM 4 Mike Jones left and Anil Kumar do some hands on work for the hardware side The group identifies optimization opportunities and default performance for hardware and software configurations p results in better performance and ease of development and deployment Java Magazine You ve been Intel s representative on the JCP Executive Committee since 2011 Can you discuss what motivated Intel to join the JCP and the benefits you see for current and future technology initiatives Kumar In 2002 when Intel first joined the JCP a hardware platform was typi cally tied to just one vertical stack But Java enabled developers to deploy the same applications on different archi tectural platforms Intel joined the JCP to
29. which is an array of gar bage collection GC log records called gcLogRecords Calling parallelStream on the data source results in the work being executed as a fork join operation By default the fork join frame work uses a common thread pool and the default size of that thread pool is equal to the number of logical cores Response Time Threads and Queues The response time or latency of a request is the composite of the time when the thread is making forward progress and the time when processing is stalled known as dead time Typically dead time accounts for a significant portion of the response time and typi cally most of that time is spent in a queue Thus if we can reduce the time spent in queues we should be able to see significant reduction in latencies In the case of a fork join opera tion what we d like to know is how many threads are in the common thread pool and how many out standing requests are in the work queue It is this information combined with other resource utilization metrics that will help us determine whether our sys tem could benefit from a larger or smaller common thread pool It s easier to talk about queues if we first talk a bit about queueing theory and in particular Little s law Little s law states that the number of tasks in a system L is equal to the average amount of time a task spends in the system lambda SAZINE MM MM M MB MM PPM MTM MARCH APRIL 2
30. word we consider is shorter than four letters the streams will be empty But that s not a problem because the stream max method returns an instance of Optional which can be empty How can we merge those two streams into one We have two solutions for that Use either the concat method or the of method Because the concat call is stateful we might think it won t be as efficient in parallel as of so let s use of see Listing 10 This merged stream of letters still needs to be mapped to a stream of scores for the letters and then the maximum score needs to be deter mined Because there are words of fewer than four letters the two streams can be empty and merg ing them will also return an empty stream So we should ensure that we cope with an empty Optional correctly If we have no letter that can be placed on the DOUBLE word word chars limit 3 Function lt String IntStream gt last3 word gt word chars skip Integer max O word length 4 J Download all listings in this issue as text LETTER SCORE square this func tion should return zero meaning that no bonus will be awarded See Listing 11 We can then write our last method which computes the full score of the best first move as shown in Listing 12 We should also add another filter step to remove words longer than seven letters because seven is the maximum number of let ters allowed in a Scrabble word With this new scori
31. you ve launched JITWatch click the Sandbox button in the top left of the main window to open a Sandbox window like the one shown in Figure 1 As well as supporting Java source code the Sandbox has sup port for Scala and Groovy because these languages compile to byte code that executes on the Java Virtual Machine JVM Other JVM languages such as Kotlin Clojure JRuby and JavaScript using Nashorn will be supported in the future Behind the scenes JI T Watch uses a java lang ProcessBuilder to compile and execute your code Experimenting with Java HotSpot VM Switches Let s begin by looking at which Java HotSpot VM switches can be used to control JIT compilation Click the Configure Sandbox button to open the Sandbox Configuration window shown in Figure 2 You can set up each VM language you wish to use in the Sandbox by specifying its home directory Next select Show Disassembly to instruct the Java HotSpot VM to produce human readable assembly language from the JIT compiled native code This capability requires the hsdis binary You can override the default setting for the Tiered Compilation option which controls when code is optimized Optimization is done first quickly with the C client com piler and then sometimes again with the more advanced C2 server compiler after more runtime sta tistics have been gathered The Tiered Compilation option is dis abled by default in Java SE 7 and ORA
32. 0000109aecd60 mov amp r9 amp rsp Hs oe A es A MA ta 0x0000000109aecd64 mov amp ebx amp ebp tableswitch 117 0 76 PolymorphismTest i2init gt 451 line 27 Es Coin coin null 1 82 0x0000000109aecd66 mov Oxl8 trep trai lao change the variable maxImplementa ZEN O ds uo snl zz z o4 TUER EE x a defauit 91 OxO000080109aecd6d movabs SOmtfLLLLLL ELL LCLECC trax pan rf 2 Bimoiphie dispatch tbe 1 0x0000000109aecd77 callq x0000000109abaf60 OopMap 0 00p 8 00p 16 04 ss FF megamorphic dispatch the me IG alod 1 invokeinterface deposit 77 p final int maxImplementations 3 75 ska ai F virtual call ais NC S ae a 021 alead 2 0x0000000109aecd7c inc tebp iine 26 tor int i Us i lt 100000 i AE 4 PolymorphismTest lt init gt 98 line 26 E T 85 goto 01 x000000010 aecd e mov xiB irsp r13 CANA dr pt eet MO Bn alocada 0x0000000109aecd83 mov tebp tebx 1 013 alnad 0x000000010Saecd85 mov Oxf trsp tril 0x0000000109aecdBe mov OxlO rsp rcx iload PolymorphismTest init 840 ling jox0000000109aecd93 cmp 50x186a0 tebx 0x0000000109aecd99 jge 0x0000000109aece00 Java lang if icmpge java lang PolymorphismTest lt init gt 44 j2 7 switch i maxImplementations 0x0000000109 1ecd9 b movslq ebx rB 36 System out printin moneyBox m Figure 6 ORACLE COM JAVAMAGAZINE III MARCH APRIL 2015
33. 015 public DoublesummaryStatistics calculateSequentially List lt String gt gcLogRecords throws IOException return gcLogRecords stream maplapplicationStoppedTimePattern matcher filter Matcher find maploDouble matcher gt Double parseDouble matcher group 2 summaryStatistics j public DoublesummaryStatistics calculateParallel List lt String gt gcLogRecords throws IOException return gcLogRecords parallelStream maplapplicationStoppedTimePattern matcher filter Matcher find maplToDouble matcher gt Double parseDouble matcher group 2 summaryStatistics m Download all listings in this issue as text multiplied by the average rate of entry into the system N That is expressed as L lambda N The powerful thing about Little s law is that it allows you to treat the entire system as a black box For example I worked on a system that had a requirement for 1 500 000 transactions per hour The average latency was 300 milliseconds 1 500 000 transactions per hour equals A17 transactions per sec ond so if we peek into the system the expected number of active requests at any point in time would be 0 300 417 125 If our thread pool size was limited to 8 that would result in 8 requests making forward progress and 117 requests sitting in a work queue Assuming there is enough hardware to sup port more active requests it s very likely that this system would ben efit fr
34. 0x000000010249c66b7 mov SOxtfEELEcE 29 case 1 coin dime break 0x000000010249c6bc mov trex trbp 30 case 7 coin quarter break xODUDDODMID249c6bf mov trax rsp 131 0x000000010249c6c23 mov iri3dd 0xl0 3 2 EEPE US mov Arbx 0x18 8 34 000009001 ine ic da nop 45 Ux000000010249c64d4 callg Ox0000000 46 System out println moneyBox moneyBox By 38 110 getstatic E15 Field ml El public static void main String args i 122 invokevirtual 16 Method 0x000000010249c6d8 callq 0x0000000 140 new PolymorphismTest 125 invokevirtual 17 f f Method 141 28 invokevirtual 118 Method 142 wiil3l return vi Figure 4 was successfully inlined as shown able moneyBox on source line 7 the static variable moneyBox from Journal button in the top left of the in Figure 4 using the assembly instruction PolymorphismSDime on source TriView screen The uncommon In the Assembly pane you at OxX000000010249c693 add line 9 using the assembly instruc trap for the bimorphic inlining is can see the native code inlined SOx5 rlld tion at OXO00000010249c6dd add indicated in Figure 5 with a red dot from the deposit method on Lower down in the Assembly SOxa rlld When all three Coin implemen inner class PolymorphismSNickel pane you can see the correspond To see the exact set of optimi tations are involved the method This adds 5 to the static vari ing inlined code for adding 1O to zations performed click the JIT call to
35. 14 193 0 MS 184 2 MS Main Memor PHOTOGRAPH COURTESY OF DEVOXX the lowest level modern Table 1 Figure 1 ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 java architect a CPU m Memory 10000 1000 DO loue 1 1980 1985 1990 1995 2000 2005 2010 Figure 2 DATA IS smaller and much faster memor CPU WORKS d 2 LOADED INTO MMM module whose whole purpose is THE CACHE WITH CACHE to mitigate the performance gap Figure 3 though it might look as if memory performance is catching up the gap is still huge For our example this means that if a computer worked as we imag ined in Figure 1 it would be terribly slow The CPU would wait most of the time for the memory to deliver the next element To overcome this bottleneck processor designers added a cache between the CPU and main memory The cache is a Figure 3 shows an improved model of the CPU and memory system Programs tend to access the same data and code several times within a short period of time tem poral locality and memory access is often limited to small regions spatial locality This means that if you load all the data that you use into the cache there is a high chance that you ll need it again later And because the next time you need it the data is available in the cache the performance of your programs increases tremendously Now you might wonder if we can put faster memory between CPU and main
36. 2 Je s e IT Teo P i EE cR om at ee D Amen s aiT ra Rn L pectus mec e E LE EP r Ih Eh NEN detur gian prom ceu Ln a ua Ed min p i a aaa a ara P ha A ees Sed uen Y Poe arro E ra A V A CUu a hr E ou Eos RU Ios Adde acc a nde rua ee ok eap a FOLE A ee a ee ee ee im En Pp ae s ws BI Ep Ax cT UD pt HAE ee ropa Hiver a AO Hori tane e Die dua cq AA ETE UL mM um EE emu Eae d Ee Fi DU LL EMEN lol e a a A AE eme O re uz ci DN ur ren zr wr are iens ee ee fru el rds Hsu A AA TDi i we k m ie ui D p x A a QE TEL EE ET i E m ue 4 i a W a i a E E sd mg i e LE L E PL Ld ru ra T Li a 1 Lu gt r rf de HA ua a A T A E Lo E m ha i Rt dl at e E I I uu E E it A E 5 E d NS E pS y el s rae gt 1h jew ans E ST E E Joa aa ge ir ee ee ed p ES Seb Eo A EE Rr ARS T e JE RISE kp M Er IT 1 Sa js ZA Lt Ar Y Si E e Mal I AR ene ia n c MA uM Wiesel Sb be mee ee Si m TT I 1 E pua M eee la IIR Er arn a Lm E BEEN 37i E uu i I y A A A ne E Ean a Jut ae NE me A oe a st m A LR du Rd radix A Wi en e ue err E i te Leh ap e p ada us dr Ps ae m Li EI rm Wr uen ae JI m rr mS E h REVEAL E RU Ea tiic d nga rra ha rt a e irr y Tae flea i pea Td ge eem A Tett unde b Ro A be EL ae ee A er quu Meer LE sr ol JP Lam pS Ce at rt ra P UL a re T EB gn AP UP ee e seid aeg i LEA A ae ge ae 1 Se e dm P m amm pra a A ie rt he re 4 hp m a
37. 2014 in Tel Aviv receiving 30 submissions from the community in just two weeks This makes us immensely proud of our membership says Gabel The group is also involved in the Adopt Open DK effort Java IL members Haim Yadid and Anat Levinger developed and contrib uted mjprof the monadic JVM profiler a command line tool for analyzing series of Java stack traces Updates recommended links and other tidbits are regularly published on the Java IL Twitter channel 2 ava IL The first issue of the Java IL monthly newsletter has been published And there s much more to come According to Gabel We have high plans for 2015 S a E java nation Virtual Technology Summit Global Series p Markus Eisele is a soft ware architect devel oper and consultant He works daily with custom ers on enterprise Java projects Eisele became a Java Champion in March 2014 Java Magazine Where did you grow up Eisele In Germany between Hannover and Paderborn Java Magazine When did you first become interested in computers Eisele Farly on when my parents bought me a Schneider CPC464 running CP M 2 0 and MARKUS EISELE Basic 1 0 with a cas sette recorder for stor age One of my first pro grams was a patient and appointment registry for a physical therapy prac tice Thousands of lines of code and GOTOs Java Magazine What was your first profes sional programming job Eisele
38. CLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 600 Sandbox Configuration Compile and Runtime Classpath Users chrisnewland jitwatch sandbox classes Add File s Add Folder Remove Configure VM Languages Fi Groovy ELS Java Script Kol Scala w Show Disassembly AT amp T syntax Tiered Compilation VM Default Always Never Compressed Oops VM Default Always Never Background JIT VM Default Always Never Disable Inlining FreginlineSize MaxinlineSize 325 Compile Threshold 10000 Extra VM switches Xmx1024m Cancel Save Figure 2 enabled in Java SE 8 Java HotSpot VM manages a pointer to an object called an oop ordinary object pointer that is usually the same size as the native machine pointer So on a 64 bit system the required heap space will be larger than on a 32 bit system Java HotSpot VM is able to save heap space by representing 64 bit pointers as 32 bit offsets from a 64 bit base this is known as compressed oops Disabling the Compressed Oops option simplifies the inspec tion of the disassembled native code because it eliminates the pointer arithmetic code needed to support compressed oops Make sure you understand the implications before altering this setting in a production environ ment For more information see Compressed Oops java architect The FregInlineSize ana MaxInlineSize fields control the bytecode size thresholds up to which Java HotSpot VM wil
39. Coast and Rwanda java nation FEATURED JAVA SPECIFICATION REQUEST SIP Servlet API JSR 359 SIP Servlet 2 0 is all about advancing the state of Session Initiation Protocol SIP in the Java ecosystem JSR 116 v1 0 JSR 359 v2 0 SIP is a communications protocol for signaling and i y JSR 289 v1 1 controlling communica http sipservlet spec java net tions sessions over Internet y O Protocol IP networks 3 J E Oracle s Binod PG the Specification Lead for JSR 359 describes the objectives of this JSR as improving the RFC coverage of SIP specification SIP outbound modernizing programming using POJOs better alignment with Java EE CDI latest servlet specification SIP over websockets extensibility better concurrency control updates to B2B and so on The JSR 359 Expert Group includes representatives from small and large companies including AT amp T Ericsson AB IBM Cisco Systems Thrupoint OpenCloud and TeleStax all of which are seeking greater alignment between SIP servlets and Java EE If you d like to follow the progress comment on the work or contribute new suggestions visit the JSR 359 project at Java net There you ll find a variety of ways you can share your ideas and get involved ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 gt Z O O YOUNG DEVELOPER Hania Guiagoussou Guiagoussou California high school student holds her pri
40. Figure 1 This interpretation is loop changes every element world And sometimes the useful for understanding the the second loop modifies only effects of processes at the functionality of the loops but every sixteenth element hardware level have a notice it s not what really happens How much faster will the able effect on our programs inside a computer second loop be compared to This article takes a look at Figure 2 shows a graph the first Take a guess how modern computers of relative performance The surprising answer is work at the lowest level and improvements that CPUs that if the code is executed on explores the things that can and memory went through in a typical laptop both loops affect performance recent decades Memory per take the same amount of formance improved steadily time Table 1 shows measure Instructions during the whole period but ments from three computers Ifyou try to imagine how the that was nothing compared The difference is negligible loops in Listing 1 are exe to the improvements in CPU T The second loop does only cuted your initial interpreta speed especially during the a fraction of the work so how is it possible that the first loop runs as fast as the second To understand this 17 4980HQ 2 8 GHZ MAC OS X YOSEMITE 30 4 MS 29 7 MS behavior you have to con 17 3770 3 4 GHZ LINUX MINT 14 25 8 MS 26 1 MS sider how the CPU and the M aU re Ee EU memory syste work On 17200 2 GHZ LINUX MINT
41. GALORE AND HYDERABAD INDIA Thirty thousand developers ORACLE COM JAVAMAGAZINE HUM MARCH APRIL 2015 attend this five day conference held in Bangalore on April 21 through 24 and in Hyderabad on April 25 The main GIDS activities in Bangalore focus on cloud mobile dynamic lan guages and analytics The one day GIDS MINI conference in Hyderabad offers sessions that reflect the main conference The dedicated Java track focuses on dynamic languages such as Scala Clojure Groovy and JRuby which run on the JVM GeeCON 2015 MAY 13 15 KRAKOW POLAND The three day GeeCON 2015 focuses on Java and JVM based technologies dynamic lan guages enterprise architec tures patterns distributed computing software craftsman ship mobile and much more JEEConf 2015 MAY 22 23 KIEV UKRAINE JEEConf is the largest devel oper conference in Ukraine The annual conference focuses on Java technologies for application development Devoxx UK JUNE 17 19 LONDON ENGLAND In its third year Devoxx UK is organized by the local Java com munity Its focus is on Java web mobile and JVM languages The conference is run by top devel oper talent community groups and expert event specialists avaOne Brazil JUNE 23 25 SAO PAULO BRAZIL JavaOne Brazil is the learning opportunity for everything Java in Latin America from Java 8 and the new features of Java 9 to the Internet of Things JVM languages and more
42. House of Cards it was comfortable with the big investment because David Fincher the showrunner and Kevin Spacey its star were very popular on Netflix The bread and butter of Netflix involves using data to recommend Chris Berry foreground and Andy Glover consider options Netflix uses its data to determine what content it should buy or produce c O O lt E lt x EP shows that customers might like Glover says Everyone at Netflix plays a role in either capturing that data or making use of that data Our business units can make changes quickly partly because the entire infrastructure is built on Java GROUNDED IN JAVA Netflix Open Connect includes Java pro cesses that are installed in third party data centers to govern how content is exchanged cached and streamed According to Glover this SOA includes thousands of instances that communi cate with one another to execute busi ness processes As Sun Microsystems used to say The network is the com puter It s all commodity hardware running Linux Ubuntu and Java EE 7 If a node fails we can bring up a new one easily While the infrastructure runs on Linux computers video encoding is often done on Windows machines facilitated by a Java based framework Java s easy portability among plat forms permits flexibility now and in S a From left to right Cameron Fieber Andy Glover and Zan Thrash analyze results Netflix uses
43. Java enterprise appli cations in a continuously test driven fashion With this hands on guide authors and JBoss project leaders Rubinger and Knutsen show you how to build high level components from persistent stor age to the user interface using the Arquillian testing platform and other JBoss projects and tools ORACLE ORACLE ES CLOUD ENGINEERED Reach More than 800 000 Oracle Customers with Oracle Publishing Group Connect with the Audience that Matters Most to Your Business Oracle Magazine The Largest IT Publication in the World Circulation 500 000 Audience IT Managers DBAs Programmers and Developers PROFITE Profit Business Insight for Enterprise Class Business Leaders to Help Them Build a Better Business Using Oracle Technology Circulation 100 000 Audience Top Executives and Line of Business Managers Lay z N lt oS x Java Magazine Java The Essential Source on Java Technology the Java Programming Language and Java Based Applications Circulation 200 000 and Growing Steady Audience Corporate and Independent Java Developers Programmers and Architects Copyright O 2015 Oracle and or its affiliates All rights reserved Oracle and Java are registered trademarks of Oracle and or its affiliates Other names may be trademarks of their respective owners Smart Greenhouse loT Developer Challenge winners Vladimir Redzhepov left and Dzmitry Yasevich pact but full fled
44. Meet Oracle and community experts in a fun atmosphere gt E O O JAVA IN ACTION JAVA TECH U D D O Da lt L java nation JAVA BOOKS JAVA 8 IN ACTION LAMBDAS STREAMS AND FUNCTIONAL STYLE PROGRAMMING By Raoul Gabriel Urma Mario Fusco and Alan Mycroft Manning Publications August 2014 Java 8 in Action a guide to the new features of Java 8 shows you how to write more concise code in less time and automatically benefit from multicore archi tectures It introduces lambdas using real world Java code It covers the new Stream API and shows how you can use it to make collection based code easier to understand and maintain It also explains other major Java 8 fea tures including default methods Optional CompletableFuture and the new Date and Time API Written for programmers familiar with Java and basic object oriented programming it demonstrates how Java 8 is a game changer diva EE 7 Developmene wih Heifsene 3 JAVA EE 7 DEVELOPMENT WITH NETBEANS 8 By David R Heffelfinger Packt Publishing January 2015 Learn how you can use the fea tures of NetBeans to acceler ate your development of Java EE applications Covering the lat est versions of Java EE APIs such as JavaServer Faces JSF 2 2 Enterprise JavaBeans EJB 3 2 Java Persistence API JPA 2 1 Contexts and Dependency Injection 1 1 and JAX RS 2 0 Java EE 7 Development wi
45. ORACLE COM JAVAMAGAZINE III MARCH APRIL 2015 ABOUT US JAVA TECH JAVA IN ACTION COMMUNITY T enterprise java option and then choose Run The Profiler window opens again pro viding the ability to view telemetry information and take snapshots this time of application memory In the snapshot view the live class objects are sorted in the Live Bytes 76 column in descending order making it easy to see which classes use the most memory Right clicking a class object pro vides the option to go to the source code see Figure 5 Now that you understand the profiling options open a browser and navigate to the AcmeWorld application Click the first example in the application which is Current Reservation List This invokes the findAll method of the ParkReservationFacade class see Listing 1 After the page loads go back into NetBeans and click the Live Results option within the Profiler window This opens a live results editor pane which displays the active hot spot methods You should see that the AbstractFacade findAll method is on top because the ParkReservationFacade APP DEVELOPMENT class extends AbstractFacade The NetBeans profiler can be a big help for locating performance problems within an enterprise application before production deployment This article just scratches the surface of what you can do with it Using Apache JMeter One of the most problematic performance
46. Oracle Java Cloud Maurice Naftalin l Faecescvei by Sri igit P E al Maurice Naftalin Effectively use lambda expressions to maximize performance improvements provided by multicore hardware QOraclePress f OraclePress www OraclePressBooks com Available in print and eBook format Netflix powers through 2 billion utting members in control and serv content requests per day with ing up targeted content based on Java driven architecture their preferences has made Netflix the world s leading internet television network Just about everyone is familiar with its flex ible on demand model Some consumers ain HMM PM MM MM MM V 1 1 11111111111 MARCH APRIL 2015 LWAYS ON are even ditching their traditional satellite and cable services in favor of an all you can eat TV service For about USSIO per month Netflix members can watch as many movies and shows as they want anytime anywhere on nearly any internet connected Java and Performance From left to right Rob Fletcher Clay McCoy Andy Glover Adam Jordens Brian Johnson and Shelley Bower are part of the always on Netflix team PHOTOGRAPHY BY BOB ADLER GETTY IMAGES 21 NETFLIX netflix com Headquarters Los Gatos California Revenue USS4 4 billion Employees 2 300 Java technologies used Java EE 6 Java EE 7 and Java EE 8 c O O lt E lt x Andy Glover and Jon Schneider left check p
47. SON JAVA TECH ith tools such as Apache Hadoop Apache Hive Apache Pig Apache HBase and others Java has become the driving force behind the big data revolution As a Java developer are you prepared to be part of this new generation of Java applications Do you have a million questions and nobody to ask See how these big data experts answered the best questions posed by the audience during the Big Data and Java Ash the Experts panel at JavaOne 2014 Fabiane Nardon a Duke s Choice Award winner is the chief scientist at Tail Target where she architects new disruptive data science services Fernando Babadopulos an expert in designing high performance software is the CTO at Tail Target Daniel Templeton has focused on big data since 2009 and works at Cloudera build Comparing Hadoop to a traditional relational system Daniel Templeton top says There ing its developer and data science Cloudera Certified might be more differences than similarities Fabiane Nardon bottom The choice of Professional certifications Simon Elliston Ball a solutions enterprise ready NoSQL database you need depends on the problem you want to solve PHOTOGRAPHS BY BOB ADLER GETTY IMAGES AND PAULO FRIDMAN ORACLE COM JAVAMAGAZINE Z444 MARCH APRIL 2015 Big data can t be processed by traditional data processing techniques says Java Champion Chris Richardson engineer at Hortonworhs helps cus tomers solve
48. a proprietary distributed file system called HDFS as well as in cloud based storage systems for example Amazon s Simple Storage Service S3 The largest amount of data Hadoop can handle is limited by the amount of data those file systems can store Because HDFS is a distributed file sys tem the cluster can be scaled up to support more data as needed A DataNode can also use more than one disk to optimize and increase storage The storage capacity of HDFS clusters can be increased either by adding a new DataNode to the cluster or by adding disk volumes to existing DataNodes Question How does Apache Hadoop differ from a traditional relational system in terms of infrastructure Templeton There might be more differences than similari ties The following are the two most significant infrastructure differences Hadoop is built on an inher number of inexpensive unreliable nodes and so the entire system is built around redundancy and fault tolerance The typical relational sys tem is built as a centralized system that runs on a small number of pow erful and reliable nodes In Hadoop the data is completely decoupled from the metadata or table structure This approach is called schema on read For example you can upload data into HDFS and then create a table over that data in Apache Hive The table metadata exists only in the Hive metastore the data itself is unaffected If you m TON pis complex and gettin
49. aFX became mature right on time for being part of the mobile revolution Native applications on mobile devices are becoming an interesting area of development The porting efforts for providing JavaFX on mobile devices are a Java community initiative and they still require lots of work To get involved in the porting efforts send an e mail to info javafxports org The port of JavaFX to Android is a continuous effort The build process has been simplified since this article was written and you can now build iOS applications leveraging the same code and tools Read more about this effort here We wrote a Gradle plugin that runs your existing JavaFX code on desktop Android and iOS devices Using the new javafx mobile plugin is very simple it s described here axticle LEARN MORE e JavaFXPorts website e JavaFXPorts sample applications sl E Hint Review a presentation on the Date and Time API Dad LJ fix this T tie November December 2014 ISSUE d h e Abuishek Gupta sLoWed us A problem USING iue a JAX RS API The correct answer is 3 Because the ChildResource class overrides the Consumes Annotation on the greet Method one Would have fo include all the other metadata annotations that ave there on the greet method in the ParentResource class In this case it happens lo be the APOST annotation This issue s Challenge comes from Stephen Chin Oracle Java Technology Ambassa
50. acilitate collabo ration and participation among both current and new JCP members Java Magazine How can the JCP better serve the Java community Kumar Several JSRs pertaining to the cloud and big data have been func Anil Kumar left and Soji Denloye discuss issues on the fly Increasingly debate in the Java community is about the Internet of Things loT and how to balance innovation and standards tioning well there s one on concur rency JSR 166 and others are related to annotation JSRs 269 and 308 and management JSRs 73 77 88 352 and so on which enable easier deploy ment But similar approaches and JSRs in the IoT space are still somewhat lacking The JCP serves primarily as a governance and approval body but an active push toward facilitating new JSRs around IoT would both enhance the perception of the JCP and better serve the Java community Java Magazine What would you like Java developers to understand about ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 the processes of the JCP Kumar Before I joined the Executive Committee my impression as a Java devel oper was that corporate representatives played the primary roles within the JCP and other such governing bodies But having now been a part of the JCP for the past three years I ve discovered that individual contributors and Java developers can also play important roles There were mechanisms in place for such contributi
51. ant But what happens before that Will the execution time be constant all the way through Figure 5 shows the dependency between array size and access time ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 4S private static final int ARRAY_CONTENT 777 Param 1024 2048 4096 8192 16384 536870912 public int size public int array public int counter public int mask OSetup Level Iteration public void setUp 1 final int elements size 4 final intindexes elements 16 mask indexes 1 array new int elements Arrays fill array ARRAY CONTENT counter O seqIndex i 16 i j Benchmark public void benchLoop 1 array 16 counter 3 counter counter 1 amp mask j J Download all listings in this issue as text As you can see there is a relation ship between these values A single modification is faster if the array is small But it s not that simple The resulting curve looks like a staircase The access time remains constant until the array size exceeds a specific threshold and then it jumps to a new level where it remains until the next threshold is reached Why is there a depen dency at all and where do these levels come from The cache is usually not a single unit but consists of several lev els with different sizes and access java architect 7 5 5 4 ns 3 2 1 O 1 L T T T T T L 1 T L T T i i L L 1 2 4
52. are events during execution I perf stat ARGS PROGRAM Without any arguments this command starts PROGRAM tracks some general events and prints out the statistics as soon as the program ends You can see a typical output in Figure 8 You can S E java architect specify which events should be tracked by using the e option To get a list of supported events run the command I perf list In Java you usually don t want to track the whole program because this would include bootstrapping the VM JIT compilation and so on which typically we re not interested in However with perf you can hook into a running process with the p option Perf is a little weird when used with the p option because you still have to specify a program that will be executed and the mea surement ends once this program ends Typically you d use the sleep command which enables you to specify the duration of the test First measure both your loops using the default settings of perf CYCLES 19 430 433 800 STALLED FRONT END CYCLES 21 361 032 STALLED BACK END CYCLES 843 462 646 Table 2 L1 CACHE LOADS 9 98 001 370 L1 CACHE MISSES 360 9 378 Table 3 This provides a good overview The most interesting results can be seen in Table 2 The number of stalled front end and back end cycles differs significantly between both runs stalled cycle means the CPU is idle and waiting for something One of the most likely
53. are the main NoSQL databases Nardon There are plenty of enterprise ready NoSQL databases out there The choice depends greatly on the prob lem you want to solve Considering the popular data model classification some good examples are Column store Apache Cassandra Apache HBase and Apache Accumulo Document store MongoDB CouchDB and RavenDB Key value store Redis Riak Amazon DynamoDB Aerospike and FoundationDB Graph store Neo4j Titan OrientDB Apache Giraph and InfiniteGraph Question What is the best way to get started with Apache Hadoop Templeton Hadoop is complex and getting started with it can be daunting However many tutorials and training options are available to help you First go buy and read a copy of Hadoop the Definitive Guide It s the single most referenced and useful text on Hadoop Online you ll find many tutorials The venerable Yahoo Hadoop Tutorial is still a good place to start You should also look at the Hands on Hadoop lab 32 BADOPULOS Because Hadoop needs to set up the job and synchronize all nodes to process the data the more data it has to process at Once the more efficient it will be PHOTOGRAPH BY PAULO FRIDMAN which was session HOL4041 at JavaOne 2014 At the time of this article Oracle had not yet posted the lab materials online The full lab is available on GitHub The lab manual includes instructions for performing
54. arkReservationFacade findCount NES 41 2 ms 94 Self time 2 59 ms 99 org javaee7 session ParkReservationFacade init 0 005 ms ES admin listener 2 A oms LJ ejb thread pooll NN 0 270 ms 6 EJ ejb thread pool2 8 072 ms LZ http listener 1 2 NENNEN 0015 ms T Ehud ain EH HotSpots X B5 Combined amp 9 Info Figure 4 J Aes Gt Class Name Live Objects Live Bytes X Live Bytes Live Object char MA 22 621 040 B 2 5 650 217 byte 0 To Source E 11 482 232 B 28 java lang Object A 6 894 984 B 135 java util HashMap Node B 5 344 832 B 167 java util HashMap Node il 4 343 080 B 46 java lang String i 4 333 824B 4 180 int i 4 165 176 B 4 7 11 java util ArrayList 3 207 216 B 3 133 java lang reflect Method I 2 394 568 B 2 7 27 java util HashMap i 2 357 040 B 2 7 49 Y Class Name Filti oa Info Figure 5 tings for enterprise components such as the Enterprise JavaBeans EJB container Java Message Service JMS and Java Transaction Service Always review the default configuration settings and modify them accordingly to support a production environment Configure server clustering to provide high availability through load balancing where appropriate ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 Do not treat your database as a black box for your data Database access can become a point of con tention for applications either when querying a large data set or when perform
55. art It can guess the next memory location correctly only if the memory loads fol low a regular pattern In the first case when you went through the array sequentially guessing the next memory location was easy and the prefetcher could mitigate a substantial part of the performance loss by prefetching the next memory location But when you accessed the array ran domly guessing the next memory location correctly was impos sible and the algorithm had to Microbenchmark Harness JMH Loap MEANWHILE FR PREFETCH cA NEXT CACHE LINE OL alada a Figure 9 wait until data was loaded from main memory The access pattern has a signifi cant influence on the performance of an algorithm but the chances to use this knowledge within Java are limited You have close to no con trol over how your data is arranged in memory But there is hope The proposed value types might provide this ability one day because they allow you to arrange object like structures sequentially in memory Another Valuable Tool Microbenchmarking is another valuable tool for getting more insight into your programs Probably the most important rule of microbenchmarking is to always use a tool that helps you to avoid some of the many pitfalls such as the insufficient warmup of the VM dead code elimina tion and loop unrolling The Java E E from Oracle is probably the best harness available right now java architect
56. ckages you need to have the Android SDK Install this SDK in a directory for example in opt android sdk You also need to install the addi tional packages that provide the latest SDK tools platform tools and build tools as explained here In addition you need to install the latest javafx for dalvik run time also called the dalvik sdk which can be downloaded from this Bitbucket page Unpack the zip file in a directory for example in opt dalvik sdk The Dalvik SDK contains a samples directory that includes a sample JavaFX HelloWorld appli cation along with the required build scripts to create an Android package You might also want to download the samples that are part of the JavaFXPorts project which show specific scenarios These samples are available at this Bitbucket page The build system Now that you have all the required tools you can create a JavaFX application This article uses the Android Gradle plugin so I recommend using the Gradle build system However the Android Gradle plugin cannot be combined with the Java Gradle plugin in a single project Because developers often want to run their applications on a desktop or a S sel tich client laptop first it s a good idea to use a multiproject Gradle setup The HelloWorld sample in the samples directory of the Dalvik SDK shows how this is structured We have a root project with three subprojects core desktop and android The cor
57. concerns for a Java EE applica tion environment is user capacity How many users can access the application at the same time This question is impossible to answer in the development environment because so many vari ables are at play Even if the development envi ronment is configured exactly the same as the production environ ment some incon sistencies between development and pro duction almost always occur User capacity testing is often a game of chance However the Apache JMeter tool can help you hone your capacity testing The JMeter tool is easy to use and it con tains useful utilities for measuring performance and load testing For more information about JIMU MARCH APRIL 2015 public List lt ParkReservation gt currentReservations return ejbFacade findAll j 2 Download all listings in this issue as text installing and using JMeter see the Apache JMeter User s Manual To begin open JMeter and create a new test plan Name the test plan AcmeWorld Testplan Next right click the test plan in the left menu and select Add gt Threads Users When the Thread Group panel is displayed enter a value of 500 into the Number of Threads field to represent 500 con current users Next add an HTTP Request to the test by right clicking the newly created thread group and selecting New gt Sampler gt HTTP Request see Figure 6 In the HTTP Request panel fill in the server name and the po
58. ction to compute the number of blanks for a word as shown in Listing 6 We can now add a filtering step in the computation of the best word built on the following predicate knowing that we have two blanks in the game Predicate String checkBlanks Word gt nBlanks apply word lt 2 Compute the score including blanks Computing the score of a word can t be conducted with the scoring method we used earlier JUH MARCH APRIL 2015 ToLongFunction lt Map Entry lt Integer Long gt gt blank entry gt Long max OL entry getValue scrabbleAvailableLetters entry getKey a 3 Download all listings in this issue as text because all the letters don t nec essarily count What we can do is once again use this histogram and multiply the score of each letter with the number of the let ter used maxed by the number of available letters Let us write a function that com putes the score of a given key value pair from this word s histogram See Listing 7 We can then use the function created in Listing 7 to compute the score of a word as shown in Listing 8 Using the new scoring function created in Listing 8 instead of the first one we created and adding the filtering step to check whether the word can be written with two blanks confirms that squeezes is the best word with 26 points The first word we found whizzing isn t far away with 23 points Place the Word on the Board But our
59. ctions for building hsdis can be found here Enter the Sandbox The Sandbox is a JITWatch feature that lets you edit code and then compile execute and analyze the Java HotSpot VM JIT logs all from within the JIT Watch appli ANALYZE THIS Understanding Java JIT Compilation with JITWatch See the effects of small source code changes and Java HotSpot VM switches cation It s intended to help developers understand what goes on under the hood of Java HotSpot VM when a program is run and to see the effects of small source code changes and Java HotSpot VM switches Note that iso lated testing of algorithms inside the Sandbox might not result in the same Java HotSpot VM compiler deci sions that would be made when running the full application That s because Java HotSpot VM will have less profil ing information on which to base its optimization decisions a 33 java architect eoo Sandbox Code Compile Execute and Analyse JIT logs Naw Editor Open Sava Configure Sandbox Reset Sandbox Java Run M SimplelnliningTest java X The Sandbox is designed to help you learn about the HotSpot JIT compilers Please note that the JIT compilers may behave differe when iting a method ff in the Sandbox compared to running your whole application public class BimpleInliningTeat public SimplelnliningTest in Late i 10 11 for int i i 1 0p i 12 13 sum this add sum Figure 1 Once
60. ddyshack Some Like it Hot Breaking Bad or any other movie or television show Netflix first executes some core Java services to determine who you are where you live if you paid your bill and what display device you are using Then its content delivery net work determines the closest cache from which to stream your content buffered to the optimum audio and video quality for your connection at that moment When a customer launches Netflix behind the scenes the system kicks off about a dozen different processes to authorize that person figure out what device he or she is using make sure the account is current and look at their recent activity explains Glover That kind of orchestration is done with Eureka an open source tool based on Java 23 Netflix stores master copies of digi tal films in Amazon Web Services and Amazon Simple Storage Service clouds Each film is encoded into more than 50 versions based on video resolution and audio quality amounting to more than a petabyte of data From there it is dis tributed to content delivery networks such as Akamai Limelight and Level 3 which in turn feed it to local internet service providers and on to viewers Netflix Fast Facts Year founded 1997 Total number of subscribers 57 4 million Number of hours per month that people spend watching Netflix More than 1 billion Percentage of viewers who binge watch shows 61 percent Bandwidth used Duri
61. descending order of its keys The last parameter is a downstream collector to specify that we want our words in a list We added an extra filtering step to make sure that we process only words that are allowed in Scrabble Extract the best key value pairs Selecting the best key value pairs can easily be done with the Stream API but we need to use a trick here because the Map interface doesn t define a stream method We can t build streams directly on maps in Java SE 8 The trick is to build that stream on the set of the key value pairs which can be built using the entrySet method call This method returns a Set Map Entry K V on which we can build a stream Because we built a TreeMap this set is in fact a NavigableSet which keeps the order of the key value pairs Displaying the three best elements becomes an easy task wordsByScore entrySet stream limit 3 or any value forEach System out println SAZINE MM MM M MB MM PPM MTM MARCH APRIL 2015 private static final int letterScores d b C d e f En h iF j k m n O p q r S I I V X y t u V W Z T332 1424135313731071 i Ie ab do doy de 10 nd Download all listinas in this issue as text At this point the best word is whizzing which is worth 33 points Not All Words Can Be Written Take a closer look at that word whizzing Can we really write that Notice that it contains two occur rences of the letter z b
62. dor And Java ne Content CLAY 1 THE PROBLEM The new Date and Time APL introduced in Java 3 is a ISR 310 not only introduces a new APT that is simpler And less error prone but it Also introduces new functionality to better handle time Zones And unit testing of complicated date logic One of the new capabilities of the Java 8 Date and Time APT is the concept of TemporalAdiusters Which Can be used to externalize date Adjustments Built in Temporal Adjusters handle many Common operations Suck As finding the last day of tue month or year But what jt you want lo do something specific lo your Application such As finding the next occurrence of Friday tue 1246 an unlucky day in Western superstition ART BY I HUA CHEN dramatic improvement over the original functionality in the JDK WHAT S THE FIX Look for the answer in the next issue Or submit your oWn Code challenge oM 2 THE CODE Here is a simple code snippet that defines a Temporal Adjuster lambda expression lo do exactly lus Given An arbitrary input date Lov example LocalDate or LocaiDateTime this code should return the next occurrence of Friday on the 1341 of the month TemporalAdjuster fridayl3Adjuster temporal gt temporal temporal with ChronoField DAY OF MONTH 13 while temporal get Chronor1eld DAY OF WEEK DayOfWeek FRIDAY getValue temporal temporal plus 1 ChronoUnit MONTHS j return temporal IT Examples u
63. e subproject con tains all code and shared resources for example fxml files It uses the Java plugin to build the code The desktop subproject contains no code or resources but only a simple build gradle file The con tents are shown in Listing 1 This build file specifies that we want to create an application that depends on the code in the core subproject and that we want the HelloAndroid class to be the class that starts the application While this desktop subproject is extremely simple it is often good practice to have this subproject and to use it during development The development lifecycle on Android is slightly longer than on a desktop and during development testing the application on a desktop or a laptop is often easier The Android Gradle project The third subproject which is named android is a typical Android Gradle project Gradle is the preferred build system for Android applications and the Android Gradle plugin already contains most of the tasks and build logic required for creating native Android applications The android directory in the samples HelloWorld demo has a typical setup for an Android app A build gradle file for building the packages An AndroidManifest xml file which contains application specific information An assets directory with some property files that will be com piled into the package You don t have to change those but they should be present Note that a typical
64. e that the application depends on the code compiled in the core subproject We also have to add the JAR files provided by the Dalvik SDK see Listing 3 In the android section of the build gradle file we have to specify the location of native libraries and properties see Listing 4 The dalvikSdkLib that is used at several locations in the build gradle file should point to the rt lib directory in the Dalvik SDK for example def dalvikSdkLib opt dalvik sdk rt lib With the combination of these three projects you can now run the JavaFX application on a desktop or laptop system or on a mobile device Executing the command gradle tasks shows an overview of available tasks For example gradle desktop run executes the application on your development environment To create an Android debug appli cation and send it to an Android device that is connected to your development environment execute gradle android installDebug lo cre ate a release version for the Android application that you want to upload tich client to the Play Store you have to enter gradle android assembleRelease JavaFX on Android Features The JavaFX runtime on Android is based on the code in the Open FX source repository and it supports most of the features of the corre sponding JavaFX release for desk top platforms For example dalvik sdk 8u40 releases are based on the same codebase as JavaFX 8u40 for desktop platforms At this mom
65. eads threads threads Figure 1 There is also code to wire the MXBean into the PlatformMBean Server This exposes the MXBean to Java Management Extensions JMX clients such as jConsole or VisualVM s JMX plugin The Unit of Work The call to concurrentParallelStream in Listing 8 results in the repeated execution of two Runnable lambdas The submit method takes the Runnable and wraps it in a ForkJoinTask Inside the calculateParallel is the call sequence Files lines path parallel This will resultin a new ForkJoinTask that will in parallel process all the lines in the GC log file The two tasks are submit ted one after another and before the call to get That means they should run in parallel However the underlying lambda expressions in the calls to calculateParallel will disrupt this because they intrinsi cally make a call to get So while there will be some overlap in the execution of the calls to the fork join operation as the numbers will show the overlap is limited Getting It to Run To get all this to work we need to execute our modified versions of ForkJoinPool and ForkJoinTask that we borrowed from Open DK In OpenJDK the original implemen tations are found in rt jar and that file is loaded by the bootstrap class loader Fortunately we don t have to tear rt jar apart and rebuild it with our versions of the classes The Java Virtual Machine JVM gives us the option to prep
66. ectively replacing inter preted bytecode with faster native code using just in time JIT compilation techniques JITWatch is a free open source tool that analyzes the complex compilation log file output generated by Java HotSpot VM and helps you visualize and understand the optimization decisions it made Chris Newland devel oped JIT Watch as part of the Adopt Open DK project and itis available for download from GitHub Follow Chris on Twitter chriswhocodes for updates In Part 2 of this three part series we explored how to make Java HotSpot VM produce the informa tion JIT Watch needs In this article we walk through some of the more advanced Java HotSpot VM features and explore how you can use the JITWatch Sandbox to test the effects of code changes and configuration settings on Java HotSpot VM behavior JITWatch Refresher You can download the JIT Watch binary from GitHub or you can build it from source code using the follow ing command mvn clean install Then you can launch JITWatch using launchUl sh for Linux or Mac OS or launchUL bat for Microsoft Windows Full instruc tions for getting started are detailed in the wiki As we look at more advanced Java HotSpot VM features you will find it use ful to have the disassembly binary called hsdis installed in your Java runtime environ ment JRE so that you can view the disas sembled native code the JIT com pilers produce Instru
67. ecycle but rather treat it as a triage step when things go wrong To make matters worse Java EE performance tun ing can sometimes be much like finding a needle in a haystack The cause of the performance issue can lie just about anywhere code to a server configuration In this article you ll see some of the processes that you can use for tuning your Java EE applications pro actively so you can try to avoid scenarios like this one Approaching the Beast Proactive Tuning Performance tuning should be made part of the standard development lifecycle As such applications should be designed from the ground up with performance in mind To be forward thinking about performance means to con sider carefully all approaches for the implementation of application solutions not just the fastest approach or the one that is easiest to imple ment Java EE applications can be particularly difficult to develop around performance because several points of con tention in a Java EE environ ment can add performance burdens to an application The top performance issues experienced in Java EE appli ano The calibration was successful Click Show Details to see calibration results Warning If your computer uses dynamic CPU frequency switching please disable it and rerun calibration as changing the CPU frequency during profiling would produce inaccurate results Show Details OK from the application s
68. egin with an analysis from Kirk Pepperdine who examines when adding paral pe lelism using the fork join framework might help performance Michael Heinrichs in The Quantum Physics of Java looks at modern chip design and its effect on Java programs Chris Newland discusses the effects of small source code changes and Java HotSpot VM switches in Java just in time JIT compilation and Josh Juneau explores incorporating performance tuning into the development lifecycle Our ICP Executive Series features Anil Kumar a performance architect at Intel And check out the big data Q amp A a panel from JavaOne 2014 answers the audience s most pressing questions Finally our cover story showcases Netflix which fulfills 2 billion content requests every day Now that s what I call performance The company can do so because most of the services running in its architecture are built on Java and the JVM On a personal note I ve been filling in during the search for a new and permanent We ll review all editor in chief I want to thank the regular staff who helped me over the rough spots OS d worked especially hard to get this i t And now I m pleased t compie and worked especially hard to get this issue out And now I m pleased to announce nadan wellurne that Andrew Binstock a veteran technology journalist most recently from Dr Dobb s some messages might Journal will take over the helm of Java Magazine Welcome Andrew send us your feedback n
69. end classes to the boot classpath using the parameter Xbootclasspath p lt path gt as shown in Listing 9 COM JAVAMAGAZINE 44MM MARCH APRIL 2015 LISTINGS LISTING 9 public long concurrentParallelStream int repeat String file Name ForkJoinTask lt DoubleSummaryStatistics gt applicationTime ForkJoinTask lt DoubleSummaryStatistics gt applicationStoppedTime DoubleSummaryStatistics applicationStoppedTimeStatistics null DoubleSummaryStatistics applicationTimeStatistics null long timer System nanoTime try Path path new File fileName toPath for inti O i repeat i applicationTime ForkJoinPool commonPool submit gt new ApplicationTimeStatistics calculateParallel path applicationStoppedTime ForkJoinPool commonPool submit gt new ApplicationStoppedTimeStatistics calculateParallel path applicationTimeStatistics applicationTime get applicationStoppedTimeStatistics applicationStoppedTime get timer System nanoTime timer System out printIn n Concurrent Parallel System out printin Concurrent Stats 5 applicationTimeStatistics System out printin Stopped Stats applicationStopped TimeStatistics System out printin Combined Time client double timer 1000000 0d ms catch Throwable t t printStackTrace j return timer j nd Download all listinas in this issue as text java architect Note Class loadi
70. ent the media framework is not sup ported More features of JavaFX have been added to the Android port For example JavaFX on Android now supports JavaFX 3D New components that are intro duced in JavaFX 8 have been added including the convenient DatePicker control You can use FXML to build your Scene Although using only standard ized Java code is best some times you can benefit from using Android specific code Apart from the standard Java classes Android APIs provide a rich framework that allows developers to integrate device specific functionality such as the location providers including GPS a near field communication NFC reader and so on Including Android APIs breaks portability so use a modular approach in your application where the Android specific parts are implemented in the android subproject To access the Android APIs a hook into the Android sys tem which is offered in the JavaFXPorts libraries is required Your application is started auto matically by an instance of javafxports android FXActivity a regular Android Activity that extends the Android Context class You can access this Activity and consequently all functionality that can be reached through an Activity by calling the following FXActivity activity FXActivity getInstance As an example you can get the Android NFC adapter by using the code shown in Listing 5 We ve seen how you can use Android code in
71. eraging our lat est hardware features most effectively But the resulting standards also benefit developers and end users By enabling a single powerful implementation across multiple environments the entire eco system benefits Java Magazine At JavaOne 2014 Intel announced it was joining the OpenJDK community as a contributing member What are the motivations and benefits for this role within OpenJDK Kumar OpenJDK provides an environ ment for innovation at a rapid pace And many of the most important technology players are already part of the OpenJDK community That has created a larger and healthier ecosys tem for competition and collabora tion Intel s new features including security features networking and storage are rapidly evolving The JCP standards process understandably 16 takes time because it requires a more thorough scrutiny But at the same time OpenJDK provides a fast path to enable these hardware features in a robust manner Java Magazine Math libraries and analytic solutions are important within the big data space Is this also part of Intel s work within the Open DK community Kumar From Intel s perspective there are two basic types of performance related innovations At the hardware layer we focus on accelerators in functional areas But it s also possible to fine tune the software layer of the stack which can exploit either exis tent or new hardware features more efficiently Beca
72. experience PHOTOGRAPHY BY BOB ADLER GETTY IMAGES ORACLE COM JAVAMAGAZINE 11 1111111111111111111111111111111111111 MARCH APRIL 2015 gt E Z gt 2 2 O O JCP Executive Series INTEL AND ORACLE Performance Innovation and Success Intel s Anil Kumar describes how collaboration drives a healthier Java ecosystem BY STEVE MELOAN Software and Services Group at Intel plays an active role in the Java ecosystem He contributes to several benchmarks and standards organizations as Java chair for the Open Service Gateway OSGi specification and as Intel s rep resentative on the Java Community Process CP Executive Committee He enables better user experience and resource use through customer nil Kumar a performance architect in the 14 applications and identifies optimization opportunities and default performance for hardware and software configura tions Kumar has presented at JavaOne SouJava Brazil and the International Conference on Performance Engineering ICPE on topics related to Java optimiza tions application latency and system power consumption Java Magazine What are the coengi neering processes that Intel and Oracle have engaged in around performance within the realms of big data cloud Internet of Things IoT and computa tionally intensive applications Kumar The coengineering work between Intel and Oracle is multifac eted These areas can be diverse and they each
73. figure Monitoring Application Server Res r 73 Applications aa Lifecycle Modules lg Resources i Figure 9 Application AcmeWorld Component Faces Servlet 4 Monitor 18 Statistics Web Container Servlet Statistics AcmeWorld Last Sample Name Value Start Time Time Details Description T Jan 7 2015 Tn TotalServietsLoaded 0 count 11 57 50 PM Total number of Serviets ever loaded 16997 Jan 7 2015 Jan 8 2015 sServletProcessingTimes millisecond 11 57 50 PM 12 01 03 AM Cumulative Servlet processing times High Water gt S Jan 7 2015 Mark 0 count ActiveServletsLoaded Ocount 41 57 50PM 7 Low Water Number of Servlets loaded Mark 0 count Web Container Session Statistics AcmeWorld Web Container Request Statistics AcmeWorld Last Sample Name Value Start Time Time Details Description is Jan 7 2015 Jan B 2015 me A RequestCount 450 count 11 57 50 PM 1201 08AM Cumulative number of requests processed so far MaxTime 2264 Jan 7 2015 Jan8 2015 Longest response time for a request not a cumulative value but the largest millisecond 11 57 50 PM 12 01 06 AM response time from among the response times ErrorCount Scout Jan 7 2015 Jan B 2015 _ Cumulative value of the error count with error count representing the number of En 11 57 50 PM 12 00 54 AM cases where the response code was greater than or equal to 400 Figure 1O
74. for POST Browser compatible headers Distribution Graph alpha Open Generate Summary Results Merge Graph Results Save Selection As JSR223 Listener Save as Test Fragment Name Mailer Visualizer Encode Include Equals Save Node As Image 3G HORREN PERI Save Screen As Image 3G aspousa uma Graph Save Responses to a file Enable Detail Add Simple Data Writer Up Down Disable Spline Visualizer Toggle de T summary Report View Results in Table Help View Results Tree Parameter Mame MIME Type Add Proxy Server Server Name or IP Port Number Username Password Figure 7 ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 enterprise java only once but if you make any significant changes to the environ ment recalibrating is a good idea To calibrate choose Advanced Commands Manage Calibration Data from the Profile menu Next choose the targeted Java plat form and click the Calibrate button After the calibration is com plete a success mes sage is displayed see Figure 1 By using the profiler you can perform a myriad of profiling tasks against a NetBeans project This article focuses on the default profiling configuration but other options are available The three main options are Monitor CPU and Memory Each of those contains its own suboptions for performing quick or customized profiling Read the Introduction to Profiling Java Applications in NetBeans IDE for more information To monitor t
75. formance is critical creating your own less flexible version of a core library method can result in significant performance improvements As a test a method was writ ten that converts String objects to uppercase It works only on the ASCII characters a through z and its performance is measured against java lang String toUpper Case using the OpenJDK JMH benchmarking framework Run this benchmark on the command line Class W Source Bytecode W Assembly Compile Chain JIT Journal Source RO F import java util concurrent Timetnit import org Cu LY de te import org o Package org adoptopenjdk itwatcn DbenchmarXks penjdk 3mh annotations Benchmark il 1 _ ES om q b penjdk jmh annotationsa BenchmarkMode Hs M org adoptopenjdk jitwatch benchmarks UpperCase import org openjdk jmh annotations Mode 9 iconat B import org openjdk jmh annotations CutputTimeUnit 10 istore 9 import org openjdk jmh annotations Scopae 2 load 10 import org openjdk imh annotations State i iload 2 11 import orq openjdk jmh infra Blackhole 15 if icmpge 12 18 aload 1 13 State Scope Thread 119 aload 14 OutputTimeUnit TimeUnit SECONDS 15 BeanchmarkMode Mode Throughput wd istore 16 public class UpperCase 26 iload 17 1281 bipuak 1B private static final String SOURCE Lorem ipsum dolor sit amet JU af iemplt 19 Ea eius porro signiferumque ius iilud eligendi vim ea
76. g started with it can be daunting An easier entry point for Hadoop is a higher level project such as Apache Hive Apache Pig or Cloudera Impala says Daniel Templeton at Cloudera ently distributed model Hadoop assumes that clus ters are composed of a large S 3 ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 The Lambda Architecture is a much debated topic in the big data community at the moment says Simon Elliston Ball at Hortonworks PHOTOGRAPH BY JOHN BLY THE change your mind about the table schema you can drop the table with out affecting the data and create a new one You can even create more than one table over the same data files Relational systems are schema on write meaning that the data and the table metadata are inextricably bound together when the data is brought into the system Question What is the overhead of using Hive over Hadoop Elliston Ball Hive works by building a series of MapReduce steps from the SQL queries This is the main source of overhead Hive was designed to run very fast over large data sets as soon as it gets started but it isn t the fastest out of the gate Something else to note here is the recent progress on Hive perfor ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 mance with Apache Tez The big differ ence Tez makes is to optimize the graph of MapReduce jobs created by Hive In addition it skips an important source of overhead
77. ged com puter with 700 MHz and memory at 512 MB This system costs around 35 Click here to watch time lapse photography of the beans growing loT Developer Challenge Winner Smart Greenhouse Control the light watering temperature and humidity of your greenhouse automatically BY KEVIN FARNHAM mart Greenhouse one of three N professional category winners in the 2014 IoT Developer Challenge is an Internet of Things IoT device and application that monitors and controls a greenhouse environment The concept for Smart Greenhouse came into being after the core team Dzmitry Yasevich Pavel Vervenko and Vladimir Redzhepov attended JavaOne Russia in April 2013 There PHOTOGRAPH BY BOB ADLER GETTY IMAGES ORACLE COM JAVAMAGAZINE VM P P B M M PM M Ml ll VIVI T T BM TP B BTgMITILG the team saw presentations of a smart house various robots and other devices all controlled by Java Yasevich notes We were impressed by these solutions and had an idea to do something like that Pavel Vervenko suggested making an automated greenhouse Everyone liked the idea First the team selected the hard ware We started to use Raspberry Pi as a basis Yasevich says It is a com MARCH APRIL 2015 However early on a safety concern arose Current under high voltage passes in the greenhouse and there is an automatic watering system so it was necessary to properly consider all the aspects related to insulat
78. gt logEntries throws IOException return logEntries parallelStream map applicationStoppedTimePattern matcher filter Matcher find maploDouble matcher gt Double parseDouble matcher group 2 summaryStatistics fad Download all listings in this issue as text in an online tutorial on Oracle Technology Network Let s define some requirements and then intro duce an interface for the MXBean The ForkJoinPoolMonitorMXBean should expose the following arrival and retirement rates the total number of tasks submitted and the average time a task spends in the system If you recall Little s law tells us that the average time of a task in the system multiplied by the arrival rate equals the aver age number of tasks in the system The average queue length com bined with the number of worker threads and a set of timers can be used to calculate the average amount of dead time accumulated by tasks waiting in the queue This could lead us to adjust the size of the thread pool assuming there is enough hardware to support the reconfiguration The definition of the ForkJoinPoolMonitorMXBean is shown in Listing 6 In addition to these methods ForkJoinPoolMonitor implements submitTask ForkJoinTask and retireTask ForkJoinTask The imple mentation shown in Listing 7 makes use of a ConcurrentHashMap to track all tasks currently in the system sa E java architect 16 worker 8 worker 4 worker thr
79. han the inaugural edition I look forward to talking about the amazing open source communities that make up Red Hat s success meeting new peers and seeing their unique solutions Also I want my girls to have a great start in their new adventures elementary school for Svenja and kindergarten for Anna You can find Markus Eisele on Twitter as a myfear and you can visit his blog gt Z O O PHOTOGRAPHS COURTESY OF TDC 2014 BRAZIL GLOBALCODE AND MARKUS EISELE 08 MAGAZINE IHR MARCH APRIL 2015 EVENTS Devoxx France APRIL 8 10 PARIS FRANCE Devoxx France brings together more than a thousand Java enthusiasts This year s event has a bigger venue to accommo date the growing number of attendees The conference focuses on Java Java Virtual Machine JVM languages web technologies mobile cloud and agile development This year s theme Infinite Possibilities is about Java innovation and the Java ecosystem PHOTOGRAPH BY DENNIS JARVIS AT FLICKR j E a i P TIL IL y Pa y hie y 1r iM J h EA p pui A JAX 2015 APRIL 20 24 MAINZ GERMANY This conference covers a range of current and future oriented technologies from Java and Scala to Android and web tech nologies and agile development models as well as DevOps More than 200 international speakers present 230 talks and 20 day long sessions Great Indian Developer Summit GIDS APRIL 21 25 BAN
80. hat s as deep as we want to dive this time but if you d like to know more I recommend running the examples that come with the Sandbox and examining the opti mizations that Java HotSpot VM can make Figure 5 Mative size Bytecode size Compile time ms V Source Bytecode W Assembly Compile Chain JIT Journal ABOUT US JAVA TECH JAVA IN ACTION COMMUNITY Source Bytecode dou Assembly 3 public interface Coin void deposit 2 SELLE d dil em Fm EM 0000000109aecd2f test r9 eS 4 m 31 incu iy x0000000109aecd32 je 0x000000010 aece23 5 public static int moneyBox 0 All a actora 4 0x0000000109aecd38 mov OxB8 amp r9 amp r10 E TR d ipanst 3 0x000000010 a1ecd3c movabs 0x2592c5dbB rdi metadata PolymorphismTestj 7 public class Nickel implements Coin 35 tuua 0x0000000109aecd46 cmp i rdi amp r10 d T x 37 oor n x00000001059aecd49 jne 0x0000000105aece35 iload hs public class Dime implements Coin pu MEOS E PolymorphismTest init 840 p10 Abs iload 0 x0000000109aecd4f jmp 0xO0000001089aecd93 11 public class Quarter implements Coin 423 Lar i8 int 100000 0x0000000109aecd51 mov rcx 0xi8 rsp aload ke public PolymorphismTest 44 if icmpge 104 o A i Wem E AE E n i ETE CUN 417 iload 6 xOOG000010Saecd56 mov rcx 0x10 rap Coin nickel new Mickel S N 0x000000010 aecd5b mov amp r11 0x8 amp rsp E POI ELM CE ERA E DS Xia 0x000
81. he AcmeWorld appli cation right click the project in the Projects menu and then choose Profile which opens the profiler s main window Then choose Monitor A few options for thread monitoring and lock contention monitoring are KNOW YOUR APP disabled by default Accept the defaults and then click Run Doing so deploys and starts the applica tion and opens both a Profiler window and a VM Telemetry Overview window see Figure 2 within the NetBeans IDE The Profiler window provides the ability to take snapshots view dump heaps and more Take the opportunity to hover over the charts within the VM Telemetry Overview window to see detailed information about the heap or garbage collector at the specified time selection After you are comfortable with these options run the pro filer again by right clicking the NetBeans project and choosing Profile If the profiler is still active you are prompted to end the cur rent session and begin a new one otherwise you can stop the profiler manually by clicking Stop within the Profiler window This time choose CPU and then select the Advanced instru mented option The CPU profiler allows you to profile the perfor mance of all application classes or only selected classes That said choosing the default Quick JUH MARCH APRIL 2015 sampled option for the CPU pro filer can sometimes incur too much overhead and produce output that is not meaningful without
82. he early days of Hadoop there weren t a lot of options for good data governance and security However now governance is very much part of the mainstream data architec ture as a key tool A lot of things come under the banner of governance for example ensuring security controlling and auditing access to data managing metadata and man aging data lineage but essentially it boils down to people and policies The Hadoop platform has tools that assist with these tasks but often good gov ernance comes down to a culture of metadata awareness On the tools front each major dis tribution handles issues of governance in a different way Cloudera has a pro prietary solution Cloudera Manager and Cloudera Navigator Hortonworks tools are 100 percent open source with Apache Ranger Apache Knox Gateway and Apache Falcon Tools from tradi tional ETL vendors such as Informatica and Talend also work well to provide governance related services Question Is Hadoop enterprise ready Nardon Yes Many companies use it in production right now to run important services There is a thriving ecosystem around it Many commercial companies offer services for Hadoop such as dis tributions support hosting consulting and so on Afew of them are Cloudera Hortonworks MapR and Amazon Richardson There are also newer com panies such as Databricks lt article gt LEARN MORE e Big Data and Java Ask the Experts panel a
83. help contribute to the Java platform s growth evolution and innovation It ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 rk togeth When we wo was an opportunity for Intel to help empower our portion of the Java ecosys tem which was undergo ing major changes at the time And that process has been fruitful If you look at the past 1O to 15 years Java has played a powerful and influential role within a variety of technology ecosystems Intel has par ticipated in 14 JSRs Since 2011 it has been my opportunity to participate in defining Java standards within the JCP With the IoT and other technology evolutions on the horizon plenty of challenges and collaborative opportunities lie ahead So I m enjoy ing the process and looking forward to the future More details of Intel s ear lier participation on the JCP Executive Committee can be found here Java Magazine What other benefits do you see in working with the JCP Kumar Intel s focus goes beyond sim ply increasing compute power We have become a platform company and that DARDOS eraround standards it creates a healthier ecosystem more competition which drives innovation and which results in better performance gt E Z gt 2 2 O O means it s no longer just about the CPU Many other evolving features and elements benefit the end user Being a part of the JCP provides us a path for contributing to the evolution of the Java platform and lev
84. hich can lead to significantly better performance Access Patterns So the size of data influences per formance Does the order in which you access your data the data access pattern have an influence too You can change the previ au Sequential Access 18 4 prion MONDE S RP A LA A O LERRA SIDE ERE RP Al ARA IIA A dn ERO 14 4 12 ns 10 4 O fF A 0 e EN 1 2 4 8 16 32 64 1282565121 2 4 8 16 32 64 128250512 KB KB KB KB KB KB KB KB KB KB MB MB MB MB MB MB MB MB MB MB au Random Access Figure 7 Performance counter stats for process id 3195 4997 277608 task clock 529 context switches ous experiment slightly to find an answer Instead of simply running an index through the array create a second array that stores the access order Access the array sequentially as before for one time and then access it randomly and measure the difference You can see the code for both experiments in Listing 3 If you run the experiment with different array sizes and plot the result in a graph you get two curves as shown in Figure 7 Not surprisingly you can see the already familiar staircase pattern Both curves show similar access times on the lower two levels which cor relate to the L1 and L2 caches But on the third level the performance of the sequential access pattern is noticeably better The fourth level shows a significant difference Why perf stat p 3195 B r 40 sleep 5 40 run
85. hich means that at this moment it offers a significant subset of the Java SE 7 platform As a consequence JavaFX 8 applica tions that use Java SE 7 APIs can be created on desktop environments ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 and subsequently converted into Android or iOS applications In Figure 1 and Figure 2 it is clear that your JavaFX applica tion can be executed on desktops and laptops as well as on Android or iOS devices One restriction The Java runtime environments on Android and iOS do not sup port Java 8 specific APIs at the moment Therefore avoid using them in code that you want to run on mobile platforms Creating a JavaFX Application on Android Writing Java code is one thing deploying that code on mobile devices is another thing Android applications are created by per forming operations that convert Java classes and resources into an Android package a apk file This Android package can be uploaded to the Google Play Store where it can be downloaded by all Android users While doing development you can also attach an Android device to your development system with a USB cable and transfer the Android package to your device Next see how you can cre ate an Android package based on your JavaFX code Note that this build process is under continuous improvement See the JavaFXPorts website for the latest and sim plest build instructions Prerequisites Before you can build Android pa
86. ick here to listen to Stephen Chin right interview Kirk Pepperdine at Jfokus Improving the Observability of this technique has been mostly ignored by everyday Java developers banging out business applications That all changed with the release of Java SE 8 and the introduction of streams Under the hood parallel ized streams are converted into fork join tasks and that makes it easy for developers to add parallelism to their applications without under java architect standing the underlying mecha nism As much as adding paral lelism might help performance indiscriminate use of the fork join framework can certainly harm per formance Let s explore this further by looking at how streams work Streams A stream takes a collection as its source and similarly to an itera tor makes each element available to some process Streams have no storage and might or might not be bounded In addition streams support intermediate aka lazy and terminal aka eager operations Intermediate operations always return a new stream whereas ter minal operations produce a result All of this happens serially unless a parallel stream is asked for and that option is supported by the underlying datasource Listing 1 shows examples of sequential and parallel streams Note For hyper threaded CPUs there are two logical cores for every physical core In Listing 1 stream executes sequentially over the data source
87. imizing the Future of Java Through Collaboration highlighted advanced features such as Intel AVX AVX2 SIMD upcoming Intel TSX and java util zip CRC32 accelera tion which is another example of end users benefiting from Intel and Oracle collaboration Java Magazine Has Oracle delivered on the promise of increased transparency and openness within the JCP Kumar From my experience as an gt E Z gt 2 2 O O Executive Committee member I would say very much so Oracle and the pro gram management office delivered the transparency JSR which has had a significant impact And I ve seen the JCP program office working with the Specification Leads to ensure that JSRs adhere to the new transparency requirements That demonstrates a concerted effort toward openness not just in following the letter of the law Java Magazine What structural procedural changes could you imagine further improving or strengthening the JCP Kumar Two JSRs JSRs 348 and 358 are related to this question One to simplify the membership and contribu tion process is focused on individual members The other JSR addresses recent JSPA Java Specification Participation Agreement discussions focused on how to better manage IP intellectual property flow policy We are seeing innovations in the IoT space as well as in the traditional back end data space But the current IP flow is complex If we can streamline that process it will better f
88. ing too many small queries Whatever the case have a contact number for your DBA who might be able to create an index on a database table or per haps even incorporate application logic into the database where that makes sense Planning for Capacity and Preparing an Application Many tools can help you prepare an application for production release Profiling tools can play a key role in an application s development lifecycle Such tools can be used to forecast how an application will enterprise java perform after it s released into pro duction under a normal or heavy load of users Two such tools are the NetBeans profiler a module of the NetBeans IDE and Apache JMeter Both of these tools can provide useful information about the perfor mance of an application before it deploys to production and both are open source The NetBeans profiler makes it easy to learn important runtime information about your application by providing the ability to monitor thread states CPU performance and memory usage The Apache JMeter desktop application simu lates a heavy load on an application before it goes into production To try out these useful tools experiment by profiling a sim ple Java EE application named AcmeWorld You can download the AcmeWorld application from GitHub and then use it to fol low along as you read the rest of this article Using the NetBeans profiler The NetBeans profiler is built i
89. ion Yasevich says As the software was being imple mented a hardware driver difficulty arose We had a problem with Pi4 because it does not have the necessary API to work with the DHT11 humidity sensor says Yasevich To solve this the team used JNI C writing a custom driver for the sensor More people became involved including intelligent hardware expert Vasili Slapik and Vladimir Redzhepov who was instrumental in making an open source Linux based distribu tion available for Smart Greenhouse Yasevich says Any farmer can down load this distribution buy a Raspberry Pi and the necessary sensors and get their own smart greenhouse into operation Visit the Smart Greenhouse Dropbox and download the source code at BitBucket FAST FACTS s Smart Greenhouse automates green houses controlling light and watering monitor ing temperature and humidity and providing live photography The hardware consists of a Raspberry Pi sen sors an Arduino expan sion board power supply SD card cooler and web camera The backing software is composed of Oracle Java SE Embedded JNI C Pi4J servlets and a Jetty container with an HTML JavaScript front end Watch the video 12 F a C ZEROTURNAROUND JRebel RELOAD CODE CHANGES INSTANTLY ALL Anil Kumar left and Mike Jones of the Intel Software and Services Group analyze a problem together The goal is to enable a better user
90. ithin Java and in the Java ecosystem It s solid It s not going anywhere That s why there is so much infrastructure being built on Java such as Hadoop Java provides both the flexibility and scalability Netflix needs Our technol ogy infrastructure is not a house of cards he concludes In order to con tinue to be the number one streaming video content provider on the internet we have to continually innovate Java is the bedrock of our software architec ture The core operational infrastructure of Netflix including the monitoring tools and platform services has all been built in Java lt article gt c O O lt E lt x MORE ON TOPIC Java and Performance Based in Santa Barbara California David Baum writes about innovative businesses and emerging technologies java architect Java and Performance PHOTOGRAPH BY ARPAD KURUCZ GETTY IMAGES Fork Join Operations How to determine when adding parallelism might help performance he introduction of lamb das and streams in Java SE 8 allows Java developers to easily add parallelism to their applications But will making an application more parallel actually improve its performance This is the sort of question that we at jClarity face on a regular basis To answer that ques tion let s take a look at how ForkJoinPool the common thread pool introduced in Java SE 7 is configured and how itis being utilized Le
91. java architect JarScan Tool The JITWatch download includes a tool called JarScan that will scan a list of JAR files and count the byte code size of every method and con structor it finds The purpose of this tool is to highlight methods that have more bytes of bytecode than the Java HotSpot VM threshold for inlining hot methods so you can get hints on where to benchmark to see whether decomposing code into smaller units can increase the program s performance The JIT compiler determines the hotness of a method using a set of heuristics that include the meth od s call frequency and the number of times back branches such as on a loop are taken There are limits on what can be inlined including a limit on the bytecode size of the called method This behavior can be tuned using the Java HotSpot VM switch XX FreqInlineSize n the default hot bytecode size threshold is 325 bytes on 64 bit Linux More details on the JarScan tool can be found in the wiki Just because a method s byte code size exceeds FreglInlineSize doesn t guarantee that it s a per formance bottleneck The code might never be invoked enough times to reach the hot code limit Remember that JarScan is a static analysis tool and has no knowl edge of how methods are used We will now explore using the JarScan tool to examine the rt jar file which contains the core Java libraries such as java lang and java util from the latest version of
92. java architect scores is easy A histogram is a simple map computed by group ing the words by their scores Thus the keys of that map are the scores of the words and the values are the words themselves regrouped into lists Are we interested in all the con tent of the histogram No What we want to see is only the best scores So we need to select the best key value pairs from that map with the best being the ones that have the greatest values for their keys A simple way of computing that is like this 1 Build the map 2 Sort the map according to its keys in descending order 3 Extract the best for example the first three key value pairs Build the map Building a map by regrouping is done with a collector Collector is a new interface in the Java SE 8 Stream API that provides a mutable reduction Mutable here means that we reduce our stream into a mutable container a map in the example shown in Listing 2 Sort the map according to its keys In Listing 2 shakespeareWords is a collection of all the words of Shakespeare from which a stream can be obtained We can regroup these words by the scores of each word The Collectors groupingBy code does just that It exists in several versions taking different sets of parameters Here we pass a supplier to build the resulting map which is implemented by a TreeMap We pass a comparator as a parameter to its construc tor to have the map sorted in the
93. l inline small and hot methods respec tively You can also choose to dis able inlining completely You might find it interesting to measure how much performance is lost when you do this Finally the Compile Threshold field sets the number of method invocations required before the C2 server compiler is invoked and the Extra VM switches field allows you to pass any additional tuning parameters Hovering over an option will dis play a tool tip that shows the exact switch passed to Java HotSpot VM JIT Method Inlining We ll now use the Sandbox to learn about how Java HotSpot VM employs a JIT technique called method inlining which is when the call to a target method is replaced with the method body itself to eliminate the invocation overhead This early step is important for the JIT compiler because it brings related code closer together before further optimizations such as loop unrolling are performed Java HotSpot VM can inline code when the method is called on an interface type and the concrete types are not known until runtime It does this by counting the number of implementations that are used during program execution When the call is observed as being monomorphic a single implementation is seen the code can be simply inlined When the call is bimorphic two implementations are observed Java HotSpot VM can inline both implementations and use a simple condition to choose between them Whe
94. ly or whether an error was trapped Any call to get prior to setCompletion being called will result in the calling thread being blocked Ideally we would like to mea sure the time from the first call to submit to the time when the job is complete This is easily done when we explicitly call the fork join operation However this technique fails with the code in Listing 4 even though the expres sion in that listing is transformed into something similar to the code shown in Listing 3 To work around this the moni toring can be placed deeper into the call stack Doing this intro duces a small amount of error in the timing but it is well below the level of noise that you ll find in any application The code in Listing 5 was lifted from OpenJDK and instrumented with a call to emit a signal when the task starts and when the task is complete Defining the ForkJoinPoolMonitorMXBean Now that we have a signal to work with we need something to ana lyze it My favored technique in Java is to introduce an MXBean A com plete discussion about MXBeans is beyond the scope of this article However it s a technique that I teach in my performance tun ing workshop and it is covered COM JAVAMAGAZINE 44MM MARCH APRIL 2015 y USTING2 LISTING3 lt LISTING 4 lt LISTING 5 lt LISTING 6 lt LISTING 7 public class ApplicationStoppedTimesStatistics public DoublesummaryStatistics calculateParallelStream List lt String
95. m E m a a tmc ul m Lm a E d d mus MEM e i m E in ES ear pn ig i a j e a S i t i QU P T L EI Ee i i A T 5 TJ hol E dls e it 2 p z h A j ri J De Se m Ea 5 z p T Law J a aa i F E LH ia E V ca eee A A A ER E E o eT Fic REED I gre ts am apE TR x r un LE a e f Pe ixl Ya im 1 Hur ami A T E FERE CERES LU a ne RUP ka xu DR Ifa E T men f au LU uta Mert t icm UM rim d AE ig Er mms Fi ro li uer n 5 d ie al e iem PUR o T ps i m 1 1 7 j j E d Lun m y ET zm Ur UE L1 Nr te aed el ed ut Umm hd nt M xr T A e H ind E J de Seay ee T P s I ir v m cm T a nl Li i 3 Te lA i Pace ERI VE or qe Fj eddie EU lated AA eee E ele EA e r m E a a uc a a rc uel EE te ON mi Java Image Enabling SDKs that Help You See the Big At first glance it may seem difficult but it s really quite simple Atalasoft s Joltimage product is a proven SDK for image enabling your Java based web applications easily Image enabling helps to add dimension to your data so you can uncover insights such as correlations and causations hidden inside your 2 dimensional documents Our SDK does the heavy lifting for you saving time money and the headaches of figuring it out yourself Backed by our highly knowledgeable amp caffeinated support engineers Joltlmage will enable your success and make the big picture so much easier to see Atalasoft
96. me active The CPU profiler enables you to take snapshots which display the runtime of active classes and methods see Figure 4 These snapshots can be compared against others to deter mine which methods are over Worked or cause contention The snapshot editor provides the option to save the snapshots view by package class or method and filter by method name You can also compare two snapshots and when you re attempting to pinpoint a performance issue this profiler option can be handy There are also options to view by the call tree by hot spots or by a combination of those The NetBeans profiler also pro vides the option to profile applica tion memory Again this feature allows you to select between quick and advanced profiling For our demo begin by click ing the red stop icon in the Profiler menu to stop the current profiler task if you haven t already Next right click the project and choose Profile and then select Memory Next select the Quick sampled Monitoring Service x E C ft localhost 4848 common index jsf Home About User admin Domain domaini Server localhost GlassFish Server Open Source Edition Common Tasks Monitoring icd Domain View and manage the monitoring information for GlassFish Server instances server Admin Server ao Clusters Monitoring 1 Standalone Instances ito Name Guter Nene pantona View Monitoring Data t Eg Nodes NIA Con
97. memory why can t ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 LS 00A private static final int ARRAY SIZE 64 1024 1024 public int array new int ARRAY SIZE for inti O n array length i lt n i array i 3 j for inti O n array length i lt n i 16 array i 3 2 Download all listings in this issue as text we just make the whole memory faster There are mostly two rea sons For one main memory is a lot larger than cache and it sim ply takes more time to find the right address within 16 GB the typical size of main memory as of this writing than to find the right address within 8 KB the typical size of Level 1 L1 cache However probably the more important rea son is that the electronic compo nents of cache SRAM are much more expensive than the ones used in main memory DRAM in terms of heat and space Heat and space are the limiting factors in modern chip design To exploit spatial locality the cache doesn t work with individual bytes but uses cache lines instead A cache line is an adjacent part of the memory typically 64 bytes What really happens when you iterate over the large loops from Listing 1 can be seen in Figure 4 The CPU loads a complete cache line from main memory into the cache and modifies the elements in the cache directly The first loop modifies all elements in the cache line while the second loop modifies only one element 16 integers
98. memory and then takes a deep annotated with Benchmark The dive into the inner workings of a test can be configured with anno modern CPU article tations at the class level Table 4 shows the most important anno tations and their meaning Conclusion Most of the time processes at the MORE ON TOPIC hardware level have no significant effect on programs but some Java and o Performance times they do Therefore it s use ful to have a rough understanding of what goes on at the hardware LEARN MORE level and to keep up with the latest e What Every Programmer Should developments Besides being use ea eee ful this knowledge is also fascinat A ing and a great way to impress your Igor Ostrovsky Bloggin S Java 1 Development Platform fellow developers Martin Thompson s blog This was the first part of a Mechanical Sympathy two part series about the quan e Linux kernel profiling with perf en TAN an MM tum physics of Java This part tutorial MINIT ITI MARCH APRIL 2015 java architect JOS PAUMARD PHOTOGRAPH COURTESY OF THE BELGIUM JAVA USER GROUP BEJUG Shakespeare Plays Scrabble New ways of solving problems with the Stream API in Java SE 8 he Collections Framework first released with Java 2 was designed on the Iterator patterns among others and is at the heart of all our applications With the inclu sion of generics in Java 5 the Collections Framework was re
99. n the observed behavior is megamorphic more than two implementations lookup tables are required and Java HotSpot VM cannot inline the method implementations Java HotSpot VM adds an uncommon trap to the opti mized code in case the number of observed implementations increases and it has to reconsider the decisions it made Let s use the example of an inter face called Coin that declares a single method deposit and three implementations of the interface named Nickel Dime and Quarter You can observe the different behaviors when two implemen tations bimorphic dispatch and three implementations megamor phic dispatch are seen at runtime In the Sandbox open the source file PolymorphismTest java which is included in the JITWatch examples JMAJAVAMAGAZINE MIMIMMIM MM M MP MBM MEM MARCH APRIL 2015 New Editor Open Save Configure Sandbox Reset Sandbox Java Run Vig PolymorphismTest java X 1 public class PolymorphismTest 2 l 3 public interface Coin void deposit i 5 public static int moneyBox 0 E 7 public class Nickel implements Coin public void deposit moneyBox 5 E f public class Dime implements Coin public void deposit moneyBox 10 10 11 public class Quarter implements Coin public void deposit moneyBox 25 12 13 public PolymorphismTesti 14 Coin nickel new Nickel 15 Coin dime new Dime 16 Coin quarter
100. nc tion to get the score of a given letter Let us write this map ping function first It is very simple and uses the old trick of reading the correct cell of the array that holds the scores of the letters letterScores IntUnaryOperator letterScore letter gt letterScores letter a This mapping function is an IntUnaryOperator a spe cialized type of function that takes and returns an int We can then write the scor ing function using the map ping function Function lt String Integer gt score word gt word chars IntStream map letterScore sum Having the chars method added to the String class proves handy Many classes have new methods that return streams on their inter nal structure The choice has been made to represent letters by inte gers and the chars method returns an IntStream instead of a Stream Integer Three streams of primitive types IntStream LongStream and DoubleStream are cre ated for performance rea sons Using them is much more efficient than using the equivalent streams of objects because we don t have to pay the cost of boxing and unboxing their elements at each step But sometimes we need to convert this IntStream into a Stream Integer which is the case when we have to build histograms later How Shakespeare Performs Now that we can compute the score of a given word building the histogram of Shakespeare s words by their a a
101. ng peak times Netflix accounts for 33 percent of North America s downstream internet traffic i The infrastructure is based on a service oriented archi tecture SOA that handles approximately 2 billion content requests per day We have thousands of microservices running all the time explains Glover This configuration offers lots of independence to our developers as they deploy new software CONTINUOUS DELIVERY Netflix engineers constantly deliver new features to the company s worldwide user base via continuous delivery It s the cornerstone of all the company s business initiatives with engineers using continu ous delivery processes to rapidly push enhancements and bug fixes to customers at low risk and with minimal overhead According to Glover this automated pipeline enables engi neering teams to move concise feature ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 Shelley Bower Brian Johnson and Mike McGarr strategize Netflix uses NN delivery system to schedule deployments based on up to the second traffic statistics sets from development and testing into a stable production environment The Netflix software environment changes a little bit every day he adds The continuous delivery environment is the conduit for speed and innovation We don t have the luxury of downtime We can t tell customers to come back in three hours after we have completed a deployment Performance problems can cau
102. ng Amazon EC2 or Google s Compute Engine Xplenty has an intuitive interface for designing ETL dataflows and also provides Hadoop with a few clicks Microsoft s HDInsight also pro vides Hadoop as a service using the Microsoft Azure cloud platform Question What can you say about machine learning algorithms Richardson The big idea in ML is the development and use of algorithms that learn from data by constructing models that make predictions or deci sions For example a common use of ML is to make product or article recommendations based on a user s past behavior One widely used ML library is Apache Mahout The scalable algorithms implemented by Apache Mahout were originally built using Hadoop MapReduce However more recent implementations are based on the much faster Apache Spark Apache Spark also has its own ML library called MLlib A key part of many applications is a big data pipeline which extracts data from an online application into the big data system and runs ML algorithms to construct models which are then used by the online application Question What about tools for gover nance in big data Elliston Ball Forgoing governance is ORACLE COM JAVAMAGAZINE S IIIIIIILITITITITITITITITTITITITIT11111 _ MARCH APRIL 2015 The big idea in ML is the development and use of algorithms that learn from data by constructing models that make predictions or decisions says Chris Richardson no longer an option In t
103. ng function we can see that there are two best words jezebel and quickly for a score of 120 points Not bad for an opening move Conclusion The problems we solved in this article might look like trivial game problems but they solve many of the problems we face in general business applications building his tograms comparing them extract ing maximum values and han dling corner cases raised by empty ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 streams Using the Stream API we can solve these problems without much code and the solution is still efficient from a pure perfor mance point of view We used some of the new pat terns introduced in Java SE 8 but the Iterator and the if then else patterns are absent from the solu tion As we can see here Java SE8 requires us to think about the solu tions for our problems differently As a bonus we get cleaner and more efficient code The Official Scrabble Players Dictionary and the words in the works of Shakespeare as well as many other interesting data sets can be found on Robert Sedgewick s page Real World Data Sets lt article gt LEARN MORE e Stream API e Collector interface lava 8 IHE FUIURE oracle com Java m a E ORACLE S E BIG DATA AND JAVA Experts answer your questions BY FABIANE NARDON FERNANDO BABADOPULOS DANIEL TEMPLETON SIMON ELLISTON BALL AND CHRIS RICHARD
104. ng in Java SE 9 and thus these details will most likely change most of the time The only expected loss is when paral lelization puts extra pressure on the I O subsystem fork join framework soaked up that 20 percent which resulted in a corresponding reduction in response time operations are many orders of magnitude slower than memory operations These observations call for more experiments Let s mix the CPU and I O bound units of work and increase the level of concur rency We can control the number The reductions in response time don t hold when I O operations are part of the workload This is a reasonable result because I O Discussion of Results CPUs are about 20 percent idle The two units of work used in this when the lambda expression example are both big and complex implicitly uses the fork join enough that they should benefit framework Explicitly using the from parallelism The first unit of work reads all of its data from memory The second reads its data from disk Each of the units of work is run serially using parallel AE dai NR ARR EE C E DONNE I M streams and then each is run again LAMBDA SERIAL 0 61 0 0 123 using fork join operations The SIMILAR oh ooooo o060000000000lloc r oo600000600000000000000000000000000 l500 o600000600000000600000000600000000llo6 oO0000600000000000060000000000000000009000llo6 60000000000000000000000000000000d results of running the two different
105. nguage to create powerful modern JavaFX API Implementation Java Platform pem Java SE looking applications that work on desktops lap tops mobile devices and embedded systems Native OS Windows Mac OS X Linux Figure 1 Porting JavaFX to Mobile Devices One of the key benefits of Java is the write once run anywhere paradigm JavaFX is no exception to this rule If you write a JavaFX application the runtime environments handle the platform specific issues so you can focus on application specific needs On desktop systems with either Microsoft Windows Linux or Mac OS X the situa tion is as shown in Figure 1 Oracle provides JavaFX implementations for Windows Mac OS X and Linux The JavaFX porting efforts for mobile devices and embedded ARM systems are coordinated by JavaFXPorts an initiative from the Java community The codebase for tich client JavaFXPorts is based on the code for Open FX which is the official repository for the JavaFX code avail able at https wiki openjdk java net display Open FX Main One goal of JavaFXPorts is to send all the changes that are made to build JavaFX on mobile devices back to Open FX This ensures convergence rather than divergence among all implementations of JavaFX The Android platform comes with its own virtual machine named the Dalvik Virtual Machine or Dalvik for short While Dalvik is by no means a fully f
106. nto NetBeans IDE 8 so you can use it for profiling your applications out of the box To perform an accurate analysis of your applications the profiler must first be calibrated to obtain data about the targeted Java platform and environment The calibration needs to be performed Le ui i CER c A a T ih AcmeWorld Testplan Thread Group f ja WorkBench HTTP Rea T FE summary Report exi t jivee bltlt HTTP Request Name HTTP Request Comments Web Server Timeouts milliseconds Server Name or IP localhost Port Number 8080 Connect Respons HTTP Request Implementation Protocol http Method GET oc Content encoding Path AcmeWorld faces currentReservations xhtml Redirect Automatically v Follow Redirects Y Use KeepAlive M Body Data Use multipart form data for POST Browser compat Figure 6 P BC Te eo bo estpian i E Thread Group HTTP Request Y 2 FE Mo Add b Config Element P FE vies Insert Parent Timer E Summa cu SEX Pre Processors Timeouts milliseconds FF Graph Post Processors b Numb 8080 C R Y Copy 3ec fou eo umber onnect esponse Ts WorkBench Paste Sev Assertions a Aggregate Graph Du plicate TEC Aggregate Report Reset Gui ProtO Assertion Results Content encoding REMOVE sil BeanShell Listener ndo es currentReservations X per Listener Reda lly v Follow Redirects Comparison Assertion Visualizer ita
107. nvironments That made sense because most of the transactions started from those environ ments Also the mobile landscape was so fragmented that creating applications capable of running on most mobile devices was difficult and expensive Today the situation has PHOTOGRAPH BY TON HENDRIKS ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 changed End users often use mobile devices smartphones and tablets instead of laptop and desktop systems and an increasing number of busi ness transactions originate from those mobile devices On mobile devices consum ers clearly prefer native appli cations over browser based applications The mobile landscape has changed significantly over the last few years Today more than 90 percent of mobile devices use either the Google Android operating system or the Apple iOS operating sys tem Both platforms provide a convenient way for developers to distribute their applica tions through an app store The Apple App Store accepts applications created for iOS the Google Play Store is where Android applications should be uploaded The combination of a new mobile landscape and JavaFX provides a Writing JavaFX Applications for Mobile Devices With recent JavaFX ports you can now create Java client applications for mobile devices Your JavaFX Application huge opportunity to Java developers With JavaFX on mobile devices Java developers can now use their favorite la
108. om a larger thread pool Instrumenting Fork oin Operations To get the statistics needed for the type of analysis performed above we need to add some instrumenta a E java architect tion After careful examination I determined that I needed to instru ment only ForkJoinTask However the monitoring would be attached to ForkJoinPool Let s define some test cases that will provide a context for the rest of this discussion At jClarity we ve spent a lot of time looking at GC logs with Censum I took this as an opportunity to look at how lamb das might be used to parse a GC log In the following example we ll focus on records for application stopped time and application concurrent Listing 2 shows a rule for parsing application stopped times from a GC log The code in Listing 2 takes a list of GC log entries filters them by those that match the pattern and then calculates a number of sta tistics that describe the stopped time behavior We can execute this code by submitting it directly to the ForkJoinPool as shown in Listing 3 which shows the explicit execution of a fork join operation The code in Listing 3 creates a new ForkJoinTask and submits it to the common ForkJoinPool When the task has been completed a call to ForkJoinTask setCompletion is made after which the results can be retrieved The parameter to that call is a status word that lets the caller know whether the task completed normal
109. only one z Using this predicate we can now add a filtering step in the computa tion of the best word At this point the best word is squeezes which is worth 26 points What About the Blanks Thinking again though we realize that we could write the word whiz zing The game uses blank letters and we could use one for the sec ond z Using blanks has a double impact First it changes the way we filter out the words Second it changes the way we compute the scores of the words because a blank does not Write a word using blanks Computing the number of blanks needed to write a given word is also a matter of com paring the his togram we built for the given word to the score any points histogram of the available letters in the game Instead of comparing the number of letters one by one we count the number of letters that are greater than the ones available Subtracting the needed letters from the available letters does what we need If the result is negative then we don t need a blank If not then the result is the number of blanks needed for that letter All we need to do is compute which is greater this number or zero for each let ter of the given word Summing the results gives us the number of blanks needed for that word First write a function that takes a key value pair from the word s histogram and returns the num ber of blanks for a given letter see Listing 5 We can then use this fun
110. ons and the JCP has worked hard to establish even greater openness and transparency It would be wonderful to provide examples of past and current roles that Java developers have played within the JCP to explore the contri butions that have been made and how that benefited the entire Java commu nity It would serve to inspire others to become involved and to demonstrate the real world possibilities of contribut ing within the JCP Java Magazine Any parting thoughts Kumar When I look back over the past 10 years I think few could have imag ined the innovations in the hardware space mobile computing the cloud big data and IoT Looking at platform gt E Z gt 2 2 O O level innovations that have been deliv ered by Intel and Oracle I couldn t have imagined we would each be playing such critical roles in these areas It has been a fascinating decade to witness When I consider the next five years while it will be different on many levels I am certain that Oracle and Intel will continue to play major roles I want to see those roles continue along a col laborative path through a standards body such as the JCP one that is open agile and proactive Such collabora tion breeds competition and innova tion which ultimately benefits the entire technology ecosystem and the end user MORE ON TOPIC Steve Meloan is a former C UNIX soft ware developer who has covered the web and the internet for such
111. ot get a direct reply S Kay Keppler Interim Editor in Chief BIO PHOTOGRAPH BY BOB ADLER ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 Opportunities section the ultimate technology recruitment resource Place your advertisement and gain immediate access to our audience of top IT professionals worldwide including corporate and Independent developers IT managers architects and product managers wv For more information or to place your recruitment ad or listing contact tom cometa oracle com nt in front of vou X nee rre ET o P am UE C TU Fx A Ro mes Dh E aL THER LT CLE EC C AC PEOR U rjk ub un eM Ro as qo A A sd ae or apu rs dE E AA EFL 15s xn E Ee Ape Tw o SI ese iE P aec TEE tx i oa Te ae 4 E UE E um T i m um i a j H l nas p b nu Lo a oe E 2 Pn PA C i M pr T a 7 Mis i a wie i E E 2 5 m oe 4 e E i in y E EU i in ae It ate n Pg T Us j y He a E P it a aj es a rs P5 E x FU a CE a a a The answer is ri mi 3 3 ps aie Ld bn J Aids a PE r mica Dm M ed i Zz a ie La p we La d F a Aa pum zo i a ee ee ERE git E PA A E A or AAA rra mE e E Jw ue RA V zu rr M Rn DE cw ri acia du nr ar ML CL um ee ieu a L E A zr es PERS DAT TCR M Cri puro laus cn rcc Rr prr op AA dara isa ok b un ax Es SAP TT ET a o ad A el a utu A 2c Raps er Ig P I 7 n teme ug uir eT ET mer aL pP
112. ource Figure 1 enterprise java cation environments tend to be related to configuration issues and environment issues Oftentimes the applications themselves are coded fine but the application server to which they are deployed is not tuned correctly or it is config ured inappropriately for the applica tions or the intended user capacity To tune properly for a production employment perform the following steps in the order listed Application tuning Server tuning Java runtime tuning Server operating system and platform tuning Next focus on decreasing the chance of incurring perfor mance issues after a production deployment Coding for performance There are bevies of coding situations that might lead to performance overhead in a Java EE application Java EE applications are executed concurrently which can lead to bottlenecks because users might be competing for resources such as web services or databases Each remote call can add latency and processes such as serialization can be CPU intensive causing further performance degradation With these kinds of issues in mind you need to craft Java EE applications 300 MB 250 MB 200 MB 150 MB 100 MB 50 MBA 0 MBA 10 04 AM Figure 2 VM Telemetry Overview cs Lock Contention gt 10 06 A 10 05 AM Heap Size B Used Heap ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 10 04 AM Surviving
113. perform ing extended analysis Click the customize link that appears near the end of the Advanced instrumented option This link opens up the Edit Profiling Roots dialog box which enables you to choose which servlets lis teners or application packages to profile By default the Select View option is set to Web Application View allowing you to select serv lets and listeners Change the Select View option to Package View and then expand the tree menu to select only the org javaee7 concurrency org javaee7 jsf and org javaee7 session packages as shown in Figure 3 Within the Edit Profiling Roots dialog box you can also select separate classes down to the con structors and methods to pinpoint your performance measurements Click OK to close the dialog box and then ensure that the Filter option is set to Profile only project classes Because enterprise appli cations typically invoke Java classes Response Time Graph Name Response ume Graph Comments Write results ta file Read fram file Filename Response Time Graph Browse Log Display Only Errors Success setings aM 44 42 ij ru ra HTTP Request Figure 8 enterprise java that reside in the target server it is a good idea to limit the profiling to those classes that are internal to the application Lastly click Run When the CPU profiler is run ning the Profiler window and VM Telemetry Overview window beco
114. performance Measure don t guess tells us that when we tune we always want to make informed deci sions driven by solid measurements lo get the needed measure ments we often need to improve the observ ability of the application being investigated In this case we were better ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 instrumentation To simplify the task of adding instrumentation the source code for ForkJoinPool was directly modified A better way to instrument would be to use byte code manipulation through the agent framework A project is being incubated at Adopt OpenJDK on GitHub and it is hoped that it will be included in Java as a formal patch I invite you to add other measures such as steal counts already estimated by the fork join framework or the number of active threads or any other measure you think would help improve the visibility of the fork join framework lt article gt MORE ON TOPIC Java and Performance LEARN MORE M 1 ll r i i I 21 TAIR paanan i F S PIN 3 E Er y iT 0A bl java architect CHRIS NEWLAND Java and Performance CHRIS NEWLAND PHOTOGRAPH BY DAVID NEWLAND BEN EVANS PHOTOGRAPH BY JOHN BLYTHE COM JAVAMAGAZINE 44MM MARCH APRIL 2015 Part 3 O racle s Java HotSpot VM contains dynamic com pilers These compilers gather statistics from your running program that are used to optimize performance by sel
115. problems aren t quite over yet What about placing the word on the board We have two special squares not far from the central one that provide a bonus because they are marked DOUBLE LETTER SCORE Would it be possible to take those squares into account to compute the final score of our first move First notice that one of these squares is four squares away to the right of the central square So it will give a bonus to a word that contains five or more letters For a word that has between five letters and seven letters the maximum any player ever has there are three placement possibilities to the right of the central square and three other possibilities to the left of the central square So for a seven letter word optimizing the placement of the word determines the highest scoring letter from the word s first three and last three letters To compute the final score we need to find this special letter add the normal score that we previ E java architect CREATE ously computed multiply by two Function lt String IntStream first3 for the first move bonus and add 50 points if the word is seven letters long the bonus for using all our letters in one word Of course the hard part is to find the special letter The first and last three letters of a given word can be extracted with the code shown in Listing 9 These two streams hold the letter that will be double scored Note that if the
116. problems with Hadoop Chris Richardson a Java Champion is the author of POJOs in Action and the founder of the original Cloud Foundry a Java platform as a service PaaS for Amazon EC2 He now consults on microservices and works on his third startup PHOTOGRAPH BY BOB ADLER GETTY IMAGES ORACLE COM JAVAMAGAZINE MI IPM MB MM MM IB III MARCH APRIL 2015 RICHARDSON Big data isn t necessarily a large volume of data It can be data that is generated at a high velocity Big data can also be data that has a lot of variety such as unstructured data Question What is big data Richardson Big data is a somewhat nebulous term that describes data that can t be processed by traditional data processing techniques such as an RDBMS based application running on a single machine Big data isn t necessar ily a large volume of data It can be data that is generated at a high velocity Big data can also be data that has a lot of variety such as unstructured data Big data technology such as Apache Hadoop tackles the problems of vol ume and velocity by scaling horizontally using fault tolerant software which tends to be cheaper and more scalable than the more traditional approach of vertically scaling very reliable hardware Apache Hadoop deals with variety by using storage formats that support both unstructured and structured data Machine learning ML algorithms are commonly used to process big data Question What
117. r ES org javaee7 reports HJ org javaee7 rss ES org javaee7 servlet v F F F F v F F F F YF i TE org javaee 7 session Find Select View Package View Advanced Help y B Edit Projects Cancel OK Figure 3 Tuning the environment Many environmental factors can make a difference in the performance of an application Learn to love the administrative console or command line utility for your appli cation server because you spend lots of time with it For GlassFish 4 x the default domain is config ured appropriately for testing pur poses but it is likely not appropri ate for a production environment without further configuration Ensure that deployment settings are configured properly for your applications If your application server allows autodeployment or dynamic application reloading be sure to disable those settings in a production environment because they can have a significant impact on performance Also take note of how often logs are written because frequent logging can cause untimely performance issues Consider the application server configuration set G B GS view Methods Lad Call Tree Method Total Time 96 Total Time Invocatio Liconcurrent defaultManagedScheduledExecutorService managedThread NAS 43 8 ms 10 Y Y org javaee7 session EJB31 Generated ParkReservationFacade Intt Be MN m org javaee7 session P
118. r a major focus on the balance between innovation and standards Even if we begin considering these issues now it will be a one or two year time frame for the processes to play out So we hope more companies will take a role in developing new standards around the lol space i Java Magazine What recent JSRs has Intel been involved with and what were the issues for your industry initiatives Kumar We ve been contributing to sev eral JSRs One was focused on greater openness within the JCP From Intel s JIMU MARCH APRIL 2015 perspective any technology environ mentis made more robust by greater openness and competition Openness serves to speed evolution and innova tion And of course performance is also a key objective So there are relevant JSRs in the performance realm for example lambda expressions in Java JSR 335 aiding computationally inten sive tasks enabling multiple cores JSR 166 fork join framework and making the overall programming paradigm eas ier As more data is collected through sensors telemetry health devices and so on companies need to analyze itin multiple dimensions which requires tremendous compute power The way that data is vectorized can also produce manyfold benefits Intel is exploring a JSR focused on autovectorization of data where a developer codes in the traditional fashion but at the JIT just in time level we would autovectorize A talk at JavaOne 2014 Opt
119. rmance streamlining I am involved in this pro cess with data processing and analyt ics at the back end The pain points for end users are enormous data transfers large memory footprints and secu rity issues Therefore we are working COMMUNITY Anil Kumar left and Liqi Yi catch up Intel works with the Hadoop and OpenJDK communities to ensure that stacks achieve maximum performance to improve our deployments on these fronts and to maximize our solutions in future hardware designs Java Magazine What are the challenges of preserving standards while also pro Working with Hadoop Intel has a dedicated team work ing to improve the performance and the end user experience for big data users Readers can find many basic to expert level optimizations for Hadoop in these Hadoop related white papers talks and videos Optimizing Java and Apache Hadoop for Intel Architecture Accelerate Big Data Analysis with Intel Technologies Tuning Java Garbage Collection for HBase Video interviews with Eric Kaczmarek big data and Java performance architect and Soji Denloye Java performance engineer Part 1 Part 2 Part 3 and Part 4 moting future innovation How does that affect Intel and its initiatives Kumar The existent standards have worked very well up to now But discussion and debate within the community have been increasing around IoT This area is rapidly evolving and now is the time fo
120. rocesses while Glover and Adam Jordens top confer The vast majority of the services running within our architecture are built on Java and the Java Virtual Machine Glover says screen They can play pause and The vast majority of the services challenges We also have a lot of open resume watching all without commer running within our architecture are built source tools that are Java based which cials or commitments on Java and the Java Virtual Machine makes it easy to monitor upgrade and Java s innate scalability has played a JVM says Andrew Glover delivery scale our services key role in enabling Netflix to expand engineering manager at Netflix Netflix Glover has a distinguished history of its service to more than 57 million uses a stateless architecture so as we executive leadership yet his passion members in 5O countries Collectively bring in more customers we are able for writing code hasn t wavered Before these viewers enjoy more than 1 billion to bring up more instances relatively joining Netflix he held senior positions hours of shows and movies per month easily We have thousands of Java pro at Vanward Technologies Stelligent including original series documenta cesses running all the time yet as we and App47 Today he manages a team ries and feature films grow we don t have huge infrastructure of seven engineers who are responsible ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 We have thousands
121. rt num ber and provide a path to test In this case use AcmeWorld faces currentReservations xhtml for the path The JMeter test is now ready to run Start the application within your IDE Then click the Start button within JMeter to run the 500 user test on the currentReservations view You can use JMeter to read the results of the test by taking a look at different listeners Right click HTTP Request and select Add gt Listener to see a list of listeners that can be used to view various performance measures see Figure 7 The most useful graphs are the Response Time Graph shown in Figure 8 the View Results in Table and the Summary Report Putting it all together The tools and techniques discussed so far can be used together to provide a variety of simulated performance and loading tests and help you Perform code reviews to ensure that proper coding techniques are used and to help reduce costly mistakes Take an abundance of snapshots using the NetBeans profiler and save JMeter reports for compari son under different environmen tal situations Perform capacity load test ing with more threads than the number of users you expect on a normal day Use JMeter capacity testing and take NetBeans profiler snapshots in unison to measure the perfor mance of your application pro cesses under a heavy load od enterprise java GlassFish Server Open Source activatedsessionsto
122. s 4 999 CPUs utilized 0 106 K sec 0 00 0 04 100 00 100 00 0 0 19 429 96 7 08 19 006 043 778 18 296 349 545 1 386 651 837 23 934 690 23 139 3 000586757 Figure 8 0 000 0 000 3 888 CPU migrations page faults cycles stalled cycles frontend stalled cycles backend instructions 0 07 13 71 4 790 O 10 branches branch misses dL A AL OL OL HH HH seconds time elapsed ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 K sec 160 00 K sec 35 51 GHz 0 00 frontend cycles idle 0 00 backend cycles idle 60 01 insns per cycle stalled cycles per insn M sec of all branches 0 07 0 00 0 05 11 00 83 285 83 31 66 74 83 37 83 37 83 30 is the access order insignificant for small arrays but plays a major part for large arrays A Valuable Tool To get a better understanding of what is going on inside the com puter you can use the Linux pro filer tool perf which collects and prints out events generated by the CPU and the memory system while a program is executed The command line interface for perf is similar to that of git You call perf with the command you want to execute 8 perf COMMAND ARGS To get a list of all commands use perf help To get help for a specific command you can run 3 perf help COMMAND The most useful command is stat It allows perf to run another program and tracks hardw
123. sage LocalDate today LocalDate now LocalDate nextFridayl3 today adjustInto fridayl3Adjuster Something 1s Wrong with the Above Temporal Adjuster What could it be 1 Variations in month Does the code need to account for differences in month length When incrementing iue temporal 2 A time zone issue Does the code need to account for input dates with time zone components 2 An error in tue formula Maybe tie calculation of Friday lue 1341 is fundamentally flawed 4 Something else entirely IAGAZINE MTM MARCH APRIL 2015
124. se buffering errors for viewers so Netflix uses its continuous delivery system to schedule each deployment based on up to the second traffic statistics For example if Europe is in its peak view ing time the delivery team might elect to delay a deployment until off peak hours One of the core metrics that guides the business is the number of streaming starts per second contin ues Glover We watch this closely in every region If there is a dip we are alerted immediately c O O lt E lt x involves using data to recommend shows that customers might like FOLLOW THE DATA Streaming video content over the inter net isn t difficult Technically there is a low barrier to entry What differenti ates Netflix among video on demand services is all the data it collects about what customers are watching how often they use the service what choices they make before and after each show is completed which genres they pre fer and many other interrelated data points Java programs capture about a petabyte of data each day for the rec ommenaation engine which suggests ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 other shows and movies that it thinks viewers will like Netflix also uses this data to deter mine what content it should buy versus produce as well as just how to produce original content to maximize viewer ship For example when Netflix pur chased the licensing rights to
125. shuffle list for int i O i lt indexes i rndIndex i list get i 2 Download all listings in this issue as text larger 25 ns but the size is so large that the analogy falls apart 8 MB which is equivalent to 256 bottles of beer Main memory is the corner store Access time is huge 100 ns but the quantity of beer is probably more than enough for a lifetime 16 GB which is equivalent to more than half a million bottles of beer Looking at these numbers it becomes quite obvious why both loops in the initial example took the same amount of time It doesn t really matter how many sips of beer you drink if you have to run to the corner store for each bottle With the cache level hierarchy in mind take a look at the graph in Figure 5 Each plateau in the graph corresponds to a level of the cache hierarchy As long as the array fits into Ll and L2 cache access time is very low But as soon as the array becomes too large and has to be read from L3 cache access time increases noticeably And the same happens again as soon as the array does not fit into L3 cache and has to be read from main memory If you look closely you can even see the small jump between the L1 and L2 cache Size does matter Even though memory is cheaper than ever java architect before try to avoid wasting it The smaller the size of the data you use the higher the chance that it will fit into the cache w
126. t JavaOne 2014 S 1 enterprise java Improving the Performance of ja i JOSH JUNEAU Java and Performance Java EE Applications Incorporate performance tuning into your development lifecycle magine that your appli cation after months of development time and lots of testing is finally ready for production deployment The application goes live and for the first few minutes every thing seems like a success Suddenly you receive a phone call from one of the users telling you that they can t access the applica tion because it appears to load indefinitely Moments later you receive another call and yet another It seems as though your application doesn t allow concurrent access to enough users This is a nightmare scenario in which the application devel opment team must make critical performance tuning decisions under the gun and they must make each of the changes to the production environment directly ORACLE COM JAVAMAGAZINE Z N MARCH APRIL 2015 Where do you begin in this situation What could be wrong Is a server configura tion preventing enough con nections for your production environment or does the application code contain a bottleneck that causes the users to wait Many applications are not tuned for performance until after they ve been put into production Unfortunately many organizations do not deem performance tuning to be an essential part of the development lif
127. t s also take a look at how to instru ment this framework so that we can get metrics that will tell us when and how to con figure it In the March 2005 edi tion of Dr Dobb s Journal Herb Sutter uttered the now infamous phrase the free lunch is over That was fol lowed by Concurrency is E COM JAVAMAGAZINE MMIMIMIMIMIMMPM M M M MIIMIIM MARCH APRIL 2015 the next major revolution in how we write software which was followed by lots of talk about Amdahl Moore s law heat dissipation how programmers and program ming languages needed to change and so on There was also a lot of talk about read write barriers and lock free wait free algorithms as well as the use of techniques that rely on divide and conquer techniques One of the divide and conquer techniques that has gained a lot of favor is the fork join framework The Fork Join Framework The fork join framework allows you to divide a large task into a number of smaller subtasks Each of these smaller tasks is broken down even further until finally you have atomic units of work that will all be executed The results of processing the atomic units are then joined together to produce a final result Java SE 7 introduced top level support for the fork join framework with ForkJoinPool ForkJoinTask and ForkJoinWorkerThread Unfortunately the program ming idiom for using the fork join framework is fairly complex and consequently Cl
128. tal o unit count lastsampletime 1 name ActivatedSessions Total count O description Total number of sessions ever activated starttime 1420726822075 rate o current d lastsampletime 1420726823914 lowwatermark 0 name ActiveServietsLoaded description Number of Servlets loaded highwatermark 4 starttime 1420726822074 s ctivesessionscurrent o unit count current 0 lastsampletime 1 lowwatermark 0 name ActiveSessions description Number of active sessions highwatermark 0 o starttime 1420726822075 CITODCOUDIL o unit count o lastsampletime 1 o name ErrorCount c o o c o V vlc o unit count o m o 2 c o a 0000 0 0 Edition 4 1 REST Interface Figure 11 While performing a true produc tion test in a development envi ronment is nearly impossible the combination of these techniques can help you refine your application as well as your production environ ment so that less time is required for reactive tuning Taming the Beast Reactive Tuning No matter how well a system has been designed for performance something can happen to slow it down More often than not perfor ORACLE COM JAVAMAGAZINE IIT T L mance tuning is done in a reactive manner so having a plan for per forming reactive tuning is impera tive When tuning a Java EE applica tion you must peel back the onion of the stack analyzing and inves tigating each layer as the possible
129. th NetBeans 8 walks you through application development using JSF compo nent libraries such as PrimeFaces RichFaces and ICEfaces Find out how you can quickly and easily develop applications taking advan tage of JPA develop aesthetically pleasing JSF web applications and develop SOAP based and RESTful web services using NetBeans including exposing EJB functional ity as web services ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 THE JAVA TUTORIAL A SHORT COURSE ON THE BASICS 6TH EDITION By Raymond Gallardo Scott Hommel Sowmya Kannan Joni Gordon and Sharon Biocca Zakhour Addison Wesley Professional December 2014 Based on Java SE 8 this updated edition introduces the new fea tures added to the platform including lambda expressions default methods aggregate opera tions and more An accessible and practical guide for programmers of any level The Java Tutorial focuses on how to use the rich environ ment Java provides to build appli cations applets and components Expanded coverage includes chapters on the Date and Time API annotations and the latest deployment best practices Key security sections include Security in Rich Internet Applications and Guidelines for Securing Rich Internet Applications gt Z O O CONTINUOUS ENTERPRISE DEVELOPMENT IN JAVA By Andrew Lee Rubinger and Aslak Knutsen O Reilly Media March 2014 Learn a use case approach for developing
130. the lab outside of the JavaOne environment The slides used during the lab are also online For more hands on training Cloudera has the most popular set of training courses with options such as data analysis with SQL on Hadoop introduc tion to data science and Apache Spark To get your hands dirty Cloudera offers Cloudera QuickStart VM and Hortonworks offers Hortonworks Sandbox Both are VM images that include a complete single node clus ter Cloudera also offers Cloudera Live which is a live dedicated cluster hosted for free for up to two weeks in the cloud An easier entry point for Hadoop than MapReduce is a higher level proj ect such as Apache Hive Apache Pig or Cloudera Impala Because Hive and Impala use a language that is close to SQL the learning curve is often shal low Tutorials for all three are online The Hortonworks Sandbox includes Hive and Pig The Cloudera QuickStart VM and Cloudera Live include all three If you re just getting started with Hadoop you should also take a look at Apache Spark which is en route to replace MapReduce as the core compu tation engine in the Hadoop ecosystem The available resources for learning ORACLE COM JAVAMAGAZINE S IIIIIIILITITITITITITITITTITTTITIT11111 _ MARCH APRIL 2015 y h e i m Ley ed T Ten Fernando Babadopulos CTO and Fabiane Nardon chief scientist both at Tail Target EN 4 de e E m discuss how m
131. thod toUpperCaseASCII generates 69 bytes of bytecode and every method call it makes was success fully inlined highlighted in green as shown in Figure 7 Conclusion That wraps up this series on some of the optimizations that Java HotSpot VM can perform Check out the Performance Techniques section of the wiki and play with the examples that come with the JIT Watch Sandbox If you have any questions or a bug to report contact Chris through the GitHub project article MORE ON TOPIC Java and Performance a LEARN MORE e GitHub repository for ITWatch java architect Part The Quantum Physics of Java An introduction to modern chip design and its effect on Java programs MICHAEL HEINRICHS magine you have an array computers can show surpris tion might be that the array is 1990s In recent years plain Bio with 67 000 integer ele ing behavior very much like stored in main memory and CPU speed hit a limit but do ments and you run two loops quantum mechanics that the CPU reads element after not be fooled The scale in over the array as shown in seems to contradict daily element multiplies each ele Figure 2 is logarithmic Even Listing 1 Both loops multiply experience But sometimes ment by three and writes the the elements of the array by quantum mechanics has result back as you can see in three However while the first noticeable effects on our real
132. tional store you can use Apache Sqoop Alternatively you can use the Hadoop DBInputFormat component to get relational data into Hadoop If you have your data in log files you can use Apache Flume to get the data into HDFS Question Is it possible to use Apache Hadoop without needing to code in Java Templeton Absolutely There s a long list of options that let you use Hadoop without Java What follows is a quick summary of the most popular options SQL on Hadoop You now have a myriad of options to run SQL or SQL like queries against your data stored in Hadoop The two most popular are Hive and Impala both of which are Apache licensed open source projects Apache Spark Spark is the newest addition to the Hadoop family and might eventually replace MapReduce as the job execution engine Spark supports code written in Scala Python and Java Hadoop streaming Hadoop includes a package called streaming that uses a simple input output contract to let you write MapReduce jobs in any language you like In fact a common use of Hadoop streaming is to exe cute standard Linux commands such as grep cut and awk against a large data set PHOTOGRAPH BY BOB ADLER GETTY IMAGES Pig Latin Another common non Java option is Apache Pig which lets you manipulate data in a high level lan guage called Pig Latin Question What is the largest amount of data that Hadoop can handle Babadopulos Hadoop can store data inside
133. tionality and should not be relied upon in making purchasing decisions The development release and timing of any features or functionality described for Oracle s products remains at the sole discretion of Oracle Oracle and Java are registered trademarks of Oracle Corporation and or its affiliates Other names may be trademarks of their respective owners Java Magazine is published bimonthly with a free subscription price by Oracle 500 Oracle Parkway MS OPL 3C Redwood City CA 94065 1600 Digital Publishing by GTxcel ORACLE COMAJAVAMAGAZINE 12 11 11711111111111111111111111111111 MARCH APRIL 2015 et Java Certified Save up to 20 v Get noticed by hiring managers v Learn from Java experts v Join online or in the classroom Y 96 of participants recommend it v 100 report promotions raises and more ORACLE from the editor lava magazine gt E O O Find the Most Qualified Java Professionals for your Company s Future ontinually improving Java performance is more important than ever Not only does the platform continue to evolve but cloud computing magnifies the effect of a single CPU From configuring the Java Virtual Machine JVM to using best practices in your application code you have lots of ways to improve performance Introducing the Java Magazine Career and in this issue we offer several ideas i l l Java and We b
134. uch data you need to start using Hadoop and NoSQL Spark are still somewhat limited A good source of learning materials is the latest UC Berkeley AMP Camp archive Spark is also much easier to simply download and play with because it offers a stand alone mode Spark is included in the Cloudera QuickStart VM Cloudera Live and the Hortonworks Sandbox Question How much data do you need to start using Hadoop and NoSQL Babadopulos It depends on the kind of job Hadoop is running Because Hadoop needs to set up the job and synchro nize all nodes to process the data the more data it has to process at once the more efficient it will be To optimize space inside the Hadoop Distributed File System HDFS files must be bigger than the size specified by the dfs namenode fs limits min block size property Nardon With NoSQL databases many times the decision of using one depends more on the need to use a more flexible data model than on the amount of data itself For example even if you don t have a huge docu ments database it might be more convenient and natural to store your documents in a MongoDB database than in a traditional relational database because the MongoDB data model is optimized for document like structures Question How can you move data into Hadoop if you aren t using NoSQL Nardon If you use a relational database and want to move the data into HDFS process it and then maybe move the result back to the rela
135. unctional Java SE 8 implementation it has many similarities to a Java SE 7 runtime As part of the JavaFX Android port the platform specific JavaFX implementation leverages Dalvik Furthermore the native parts of the JavaFX runtime are compiled Your JavaFX Application i Java Platform JavaFX API JavaFX Android Implementation Dalvik Android Figure 2 for Android as shown in Figure 2 It is important to note that although Java SE 8 is not avail able on Android the JavaFX 8 APIs are available Also the internal implementation of JavaFX 8 on Android and iOS does not use Java SE 8 specific APIs Moreover it is possible to use lambda expressions because a great tool called RetroLambda can convert them at the byte code level RetroLambda inspects class files and it replaces all invokedynamic calls with invoke calls that are supported in Java SE 7 On iOS the situation is a bit more difficult because no Java runtime is available on the iOS plat form Worse Apple has restricted the just in time JIT options that are used by typical Java Virtual Machines JVMs and that compile bytecode to native code at runtime However RoboVM provides a smart ahead of time compiler that translates Java code to native iOS code eliminating the need for a JIT processor running on the devices RoboVM offers more or less the same base classes as the Java classes in the Dalvik Virtual Machine w
136. usage to ensure that the server is not overloaded Database monitoring might also be useful in a reactive tuning toolset Conclusion To deem an application successful it should include all of the intended functionality and perform well Users that experience poor perfor mance often consider the applica tion unworkable Incorporating proactive perfor mance tuning measures into your development lifecycle can allevi ate many hours of reactive tuning on the tail end Before production deployment it is imperative that developers have an accurate por trayal of how an application per forms for its users The NetBeans profiler and Apache JMeter are two profiling tools that can help you forecast how an application will perform under a normal or heavy load of users after it is released to production lt article gt MORE ON TOPIC Java and LEARN MORE NetBeans profiler e Apache JMeter e LightFish E of tich client JOHAN VOS a ava developers can use J JavaFX to create great cli ent applications that have a compelling user interface By leveraging the power of the Java platform developers can create client applications that address today s require ments for performance as well as look and feel This article explores the state of JavaFX on mobile devices with a specific focus on the Android platform In the past Java client applications focused on desk top and laptop e
137. use our goal is always better performance for the end user Intel strives to contribute to both the software and hardware stack evolu tions within Open DK Java Magazine At JavaOne 2014 Michael Greene Intel vice president Software and Services Group discussed Intel s work with the Hadoop commu nity to optimize hardware related to big data solutions What is Intel s vision of empowering the flow from data nibbling sensors to number crunching servers to end user services and insights What are the implications for performance demands Kumar As part of our focus on big data related libraries Intel is working with the Hadoop community and the Open DK community to ensure that stacks achieve maximum performance Garbage collection is one major area of interest And for future deployments we are working on accelerators for encryption compression and com putational analytics We ve recently published several white papers and videos that explore how end users can benefit on these fronts within the big data space See the Working with Hadoop sidebar At the CES 2015 keynote our CEO by greater openness and competition Openness serves to speed evolution and innovation MAGAZINE UNIV MARCH APRIL 2015 Brian Krzanich announced several breakthrough technologies for con sumer electronics including innova tions within the loT space There are many evolutions occurring in the IoT including our own internal perfo
138. ut only one is available in the Scrabble game How can we discard such a word What we need to do to tell whether a given word can be writ ten with the letters that are avail able in the game is to count them letter by letter and compare the number of needed letters to the number of available letters If this test fails then the word cannot be written This operation can be done with an allMatch call Counting all the letters of a given word consists of building a histogram in which the keys are the letters and the values are the number of times each let ter is used to write that word Let us first build this histogram see Listing 3 The call to the boxed method converts the IntStream returned by the chars call into a Stream Integer This is neces sary because the collect Collector method we want to use is defined on streams of objects not on streams of primary types IntStream LongStream and DoubleStream Then we need to build a stream on its key value pairs and compare those entries with the available letters in the game as shown in Listing 4 In each pair the key is the letter and the value is the num ber of times that letter is used in the word E THINK SOLUTIONS java architect The number of available letters in the game is given by reading the array scrabbleAvailableLetters declared at the beginning of Listing 4 From this array we can see that nine i s are available and
139. written but the patterns for using it weren t modified Java SE 8 introduced lambda expressions and the Stream API The Stream API brings new patterns and new ways of solving the prob lems we encounter in busi ness applications that were previously solved with the Collections Framework This article shows how you can use the Stream API to solve a simply stated problem built on the Scrabble game in just a few lines of code Shakespeare and Scrabble Let s say we have two files One contains supposealy all the words Shakespeare used in his plays and poems ORACLE COM JAVAMAGAZINE Z4 MARCH APRIL 2015 The other contains all the words in The Official Scrabble Players Dictionary The question is this How well would Shakespeare have played Scrabble Or what is the best word Shakespeare could have played and how much would it have scored It turns out that this ques tion is a map filter reduce question that can be solved with the Stream API Compute the Score of a Single Word Computing a score is about taking a given word and returning a score that is an integer This can be done with Function lt String Integer which can be implemented with a lambda expression Let s build an array letter Scores which gives the score of each letter in the English Scrabble game Listing 1 From this array we can see that the score of m is 3 and the score of y is 4 Inside this scoring function we also need a mapping fu
140. ze Hania Guiagoussou became from the SITIC A passionate Java developer nova ti n after attending a Java pro Competition gramming summer workshop at Oracle two years ago If it weren t for Alice I wouldn t be interested in programming she explains in an inter view Her most recent project a Water Saver System tracks the need for watering using humidity and temperature sensors in the soil She and her teammate won awards at a local engineering and science fair as well as third place in the Computer Science Math and Engineering category of awards sponsored by the local utility company She also won third place and US 10 000 in the Digital Innovative Challenges an African competition os java nation UO The Israel Java User Group Java IL Was formed in mid 2014 as a consoli dation of three smaller user groups HaBotzia founded by Asaf Mesika and Uri Shamay JJ TV founded by Ron Gross and Tomer Gabel and ILJUG founded by Haim Yadid Gabel notes that the idea for combining the groups originated with Haim whose efforts to bootstrap ILJUG led to the realization that com munity thrives on diversity and that each of the three groups seems to operate in its own silo In consolidating the groups we ve increased our reach and having a much larger organizational body enables more frequent diverse events The combined Java IL group had 540 mem bers at the start of 2015 Dur membership spans the entire

Java Magazine - March/April 2015

Contents

Download Pdf Manuals

Related Search

Related Contents