User Guide For Parallel Secondo
Contents
1. As shown in this example the hadoopMap operation first selects trajectories by pruning the distributed B Tree created before Result trajectories are decomposed into units which are then distributed into a global cell grid by adding a new attribute named Cell Afterwards reduce tasks fetch different interme diate results as their input based on the Cell attribute and the interFunc is processed by slaves in parallel 11 during the reduce step Note that in the above examples interFunc arguments in both hadoopMap and hadoopReduce op erations include several flist objects For each operation the first flist is set as the input delivered to the interFunc by the implicit type operator TPARA However all the remaining flist objects have to be quoted by the para operator since operators like exactmatch and product cannot accept any flist input The executed parameter in the hadoopMap is set to be false here therefore only one Hadoop job is created in the hadoopReduce operation which contains not only the Reduce stage described in hadoopReduce but also the Map stage described in hadoopMap hadoopReduce2 flist T1l x flist T2 x partAttributel x partAttribute2 subName string x subPath text kind DLO DLF x reduceTaskNum int isHDJ bool J interFunc map T1 x T2 T3 flist T3 x x x x This operator is similar as the above hadoopReduce except it accepts two flists2. it is possible to simulate a virtual cluster on one computer only and set the Parallel SECONDO up on it At present Parallel SECONDO provides at least both Ubuntu and Mac OS X platforms The deployment of Parallel SECONDO on one computer includes several steps Prerequisites Few utilities must be prepared before installing Parallel SECONDO They are all ordinary linux commands working on many Linux platforms and MacOSX Java 1 6 x or above The openjdk 6 is automatically prepared along with the installation of SECONDO SSH connection Both Data Servers and the underlying Hadoop platform rely on secure shell as the basic communication level For example on Ubuntu the SSH server is not installed by default and can be installed with the command sudo apt get install ssh screen is also requested by Parallel SECONDO scripts In Ubuntu it can be installed like sudo apt get install screen Particularly in Hadoop a passphraseless SSH connection is required The user can check and set it up through the following commands First carry out the ssh command to see wether a passphrase is required ssh lt IP gt Here the IP is the IP address of the current computer and usually can be found out through the ifconfig command If a password is asked for this connection then an authentication key pair should be created with the commands cd HOME ssh keygen t dsa P f ssh id_dsa cat ssh id_dsa
3. DLO flist PartRel is returned Assume this cluster contains two slave Data Servers then the PartRel has a 2 x 1 PS Matrix while each sub object contains half of the relation open database opt 12 let SubBTree PartRel hadoopMap createbtree PLZ let ParaResult SubBTree hadoopMap DLF para PartRel exactmatch Con_PLZ collect consume close database Here the hadoopMap is first used to create a distributed B Tree over the cluster returning the DLO flist SubBTree Afterwards another hadoopMap is used to prune the distributed B Trees in every slave The selected result is still distributed on slaves and returned as a DLF flist Here the Con_PLZ is delivered automatically to every slave Data Server although it is created only in the master database because its type ant is associated with the DELIVERABLE kind At last the collect operator accumulates all distributed sub files created by the former hadoopMap operation open database opt query PartRel hadoopMap DLF feed para PartFile n hashjoin PLZ PLZ n collect count Wrong guery should be forbidden query PartRel hadoopMaplDLF feed para PartRel feed n hashjoin PLZ PLZ n collect count query PartFile hadoopReduce Ort DLF 5 para PartFile n hashjoin PLZ PLZ n collect count query PartRel PartFile hadoopReduce2 Ort Ort DLF 5 feed n hashjoin Ort Ort n collect count close database Above
4. Parallel SECONDO one node may contain multiple SECONDO databases and a cluster may be com posed by many computers Therefore some regular routine work like starting stopping and updating SECONDO systems are better processed with some auxiliary bash scripts These auxiliary utilities are briefly introduced in Section 2 and their names are all started by ps Most of the scripts support the usage explanation which can be printed out with the h argument Here we only introduce several of them about managing the Mini SECONDO of the system common operations in Parallel SECONDO accomplished with these scripts 4 1 Update Mini Secondo The Mini SECONDO is a compact SECONDO distribution only containing essential components required to handle its database All of them are built based on the SECONDO installed on the master node keeping identical over the whole cluster In case there will be any extension made for SECONDO like creating new data types or operators the user can update all Mini Secondo in the cluster immediately cd SSECONDO_BUILD_DIR Algebras Hadoop clusterManagement ps stop AllMonitors cd SSECONDO BUILD DIR make cd SSECONDO BUILD DIR Algebras Hadoop clusterManagement ps secondo buildMini c OR OP i i OP an OP a The ps secondo buildMini provides two major arguments c cluster and 1 local If the c argument is set then the update will be distributed to every Data Server of the cluster I
5. clusterManagement ps secondo buildMini lo So far Parallel SECONDO has been installed If the user wants to completely remove it from the computer or the cluster an easy to use tool ps cluster uninstall is also provided S cd SSECONDO_BUILD_DIR Algebras Hadoop clusterManagement ps cluster uninstall B Setting Up Parallel Secondo In A Cluster Installing Parallel SECONDO on a cluster is almost the same as the installation on a single computer The auxiliary tools help the user to set up and use the system easily with only the configure file need to be adjusted Prerequisites Before the installation there are some basic environment that the cluster should prepare First all com puters in the cluster should have a same operating system or at least all of them should be Linux or Mac OSX Second a same account needs to be created on all machines This can be easily done with services like NIS Thereafter the SSH and screen services should be prepared on all computers and the comput ers can visit each other through SSH without using the passphrase Third one or several disk partitions should be prepared on every machine and the account prepared before should have read and write access on them At last all required libraries of SECONDO and the secondorc file should be installed on every computer Likewise if the operating system is Ubuntu then the line source SHOME secondorc SHOME secondo should be set at the to
6. oper ators However it is impossible to let all operators recognize and process this new data type Regarding this issue we implement the para operator to unwrap flist objects and return their embedded data types in order to pass the type checking of existing operators Note that there is no value mapping function provided for this operator since it is only prepared for letting flist objects pass through different operators type mapping functions It is designed this way as there is no generic data able to express SECONDO objects with various types Therefore the para operator can only work with the hadoop operators that will be introduced later It is set inside hadoop operators interFunc arguments which are not evaluated directly in the master database TPARA flist T T This is a type operator extracting the internal type from the input flist object and delivering it to the internal function argument as its input parameter It works similar as the above para operator but it can only be used implicitly TPARA2 ANY x flist T gt T This is also a type operator working similar like TPARA but it gets the embedded type from the second flist input instead of the first one 5 3 3 Hadoop Operators hadoopMap flist T x subName string x subPath text x kind DLO DLF x mapTaskNum int x executed bool x interFunc map T Tl flist T1 hadoopMap creates an flist object of either
7. we list four parallel queries for processing a same distributed self hash join operation with hadoopMap hadoopReduce and hadoopReduce 2 operators respectively The first query fin ishes the operation in the map step only It takes PartRel as the input flist hence each map task reads one sub object from its own Mini database and gets the other side from the PartFile quoted with the para operator If a DLF flist is used in the interFunc of a hadoop operator then all its sub files will be collected to that task Note the second query that uses PartRel in its interFunc cannot get the correct result as each map task can only get its own sub object The third query is a hadoopReduce operation the input PartFile is repartitioned into 5 columns based on its Ort attribute in its map step Averagely each reduce task finishes the interFunc with 20 data from the left side and the whole data set from the right side The last query finishes with the hadoopReduce2 operator it reads SubRel and SubObj at the same time Each map task reads the left side from its database and reads the right side from the PSFS Both side relations are repartitioned by their Ort attribute into 5 pieces In its reduce step each task gets 20 data from both sides 13 A Setting Up Parallel Secondo On A Single Node Nowadays it is common that one computer has a powerful computing and storage capability with several processors or cores and several large hard disks Therefore
8. DLO or DLF kind after processing its interFunc by slaves in parallel during the map step of the template Hadoop job Both subName and subPath are optional arguments the default flist kind is DLO and the mapTaskNum default value is the current cluster size The interFunc is expressed as a function argument and not evaluated in the master node but delivered and processed in slaves Take the creation of a distributed B Tree as an example The original sequential query is let dataSCcar_Licence_btree dataSCcar createbtree Licence This query creates a B Tree index for a relation called dataSCcar based on its Licence attribute The parallel queries are let dataSCcarList dataScar feed projectextend Licence Type Model Journey Trip spread Licence 10 TRUE hadoopMap consume let carLicence_btreeList dataSCcarList hadoopMapl dataSCcar Licence btree createbtree Licence Here both the dataSCcarList and the carLicence_btreeList are flist objects The first query distributes the tuple relation to slaves and returns a DLO flist It first uses the spread operator to distribute data as sub files and returns a DLF flist object All SECONDO indexes must be built based on stored relations where each tuple has a unique tupleID that is indispensable for index structures Therefore the returned DLF flist is sent to the hadoopM ap operator letting each slave Data Server import the local sub files to its Mini SECOND
9. Hadoop applications and the DS Catalog are only set in its first Data Server The order of these Data Servers is decided by the DS Catalog and the master Data Server is always the first one on the master node Master Node Master Data Server lt gt Meta Parallel Database Converter DS Catalog Hadoop Master Node H D F i P Slave bad Node Hadoop Slave Node S DS Catalog Partial Secondo F Slave Data Server Ll Slave Database Slave Data Slave Data 4 Server Server lt _ lt Slave Node Figure 1 The Infrastructure of Parallel Secondo 2 Hadoop Algebra SECONDO is composed by algebras each containing a set of data types and relative operators Two algebras are especially required by the Parallel SECONDO Hadoop and HadoopParallel Besides some other components are also required 1 Hadoop Archive Hadoop archive with the version 0 20 2 is used as the underlying framework for Parallel SECONDO It is not included in the SECONDO release by default and has to be downloaded to the directory SECONDO BUILD DIR bin by the user However its installation is performed automatically later along with the deployment of the Parallel SECONDO 2 Parallel SECONDO Auxiliary Utilities A set of bash scripts is provided to manage the system They are kept in the Hadoop algebra by default in a folder named clusterManagement Thes
10. O database saved as a sub object At last the second query reads the created dataSCcarList and uses a hadoopMap to let slaves create their own index by executing the interFunc 10 Limited by the current SECONDO query processor the hadoopMap only describes the Map stage in the created Hadoop job At the same time the following hadoopReduce and hadoopReduce2 operators describe only the Reduce stage in the job In order to construct jobs including both Map and Reduce stages i e both stages invoke Mini SECONDO on slaves to process the interFunc queries an additional boolean parameter named executed is provided in the hadoopMap operator By default this parameter is set to be true then the operator creates a Hadoop job with only its Map stage containing the InterFunc query Particularly when hadoopMap concatenates with hadoopReduce or hadoopReduce 2 and sets the executed parameter to be false then it returns an unexecuted flist object with the InterFunc being not processed Such an unexecuted flist is delivered to the next reduce operator in which a Hadoop job is created with Map tasks containing the InterFunc argument in the hadoopMap and the Reduce tasks processing the InterFunc described in the reduce operator An example of this feature will be explained later hadoopReduce flist T x partAttribute x subName string x subPath text x kind DLO DLF x reduceTaskNum int x interFunc map T Tl flist T1
11. RectangleAlgebra 0 Memoryblock fix core 21 string StandardAlgebra 0 Serialize variable extension 22 ubool TemporalAlgebra 0 Memoryblock fix core 23 uint TemporalAlgebra 0 Memoryblock fix core 24 upoint TemporalAlgebra 0 Memoryblock fix core 25 ureal TemporalAlgebra 0 Memoryblock fix core 26 ustring TemporalExtAlgebra 0 Memoryblock fix core 27 vertex GraphAlgebra 0 Memoryblock fix core 28 filepath BinaryFileAlgebra 1 Memoryblock fix core 29 text FTextAlgebra 1 Memoryblock fix core Table 2 DELIVERABLE data types A sub file consists of the type and data files The type file describes the schema of the exported relation being produced during the type mapping period and duplicated to every node inside the cluster Data files contain tuples binary blocks and all data files kept in one Data Server share the same type file The sub files are readable with the ffeed operator Here the PS Matrix has a scale of scaleN x scaleM They also both are optional arguments The default scaleN value is the cluster size and the default scaleM value is 1 Normally the partition attribute A is removed after the query except the argument keepAl is set as true Same for the other partition attribute AJ The AJ must be different from AJ Data files are named as fileName row column row 1 scaleN and column 1 scaleM For the purpose of fault tolerance each partition file is duplicated on dupT imes conti
12. Servers At the same time most intermediate data are exchanged among Data Servers through the PSFS A particular Data Server set on the master node is called the master Data Server its Mini SECONDO is the only entrance to the system called the master database The other databases that are deployed on slave Data Servers are called slave databases Through various PQC Parallel Query Converter operators which are also called Hadoop operators a parallel query submitted in the master database is converted to a Hadoop job and then partitioned to tasks These tasks are processed by Data Servers in parallel as scheduled by the Hadoop framework Different from the slave databases the master database contains some global and meta data of the whole system therefore it is also called the meta database as well An identical DS Catalog Data Server Catalog is duplicated on every node of the cluster describing access entries for all Data Servers It is composed by two files master and slaves listing Data Servers by lines with three elements IP_Address PSFS_Location SecondoPort For each line its first element distinguishes nodes by their IP addresses whereas the second and the third elements tell apart Data Servers within a same computer based on their PSFS node locations and Mini SECONDO ports The master file should contain only one Data Server It is possible to use the master also as a slave Data Server In a node with several Data Servers the
13. User Guide For Parallel Secondo Jiamin Lu Ralf Hartmut Giiting Databases for New Applications Fernuniversitat Hagen Germany June 3 2013 1 Parallel Secondo Infrastructure Parallel SECONDO is constructed by coupling the Hadoop framework and discrete SECONDO databases on a set of computers connected with network as shown in Figure It can be deployed on either a single computer or a cluster containing tens or even hundreds of computers Its two components Hadoop and discrete SECONDO databases coexist in the same system and each can be used independently In Hadoop nodes communicate through HDFS Hadoop Distributed File System whereas each single node SECONDO exchanges its data with the other SECONDO instances through the PSFS Parallel SECONDO File System which is a distributed file system perticularly prepared for Parallel SECONDO Whereas Hadoop is deployed by nodes Parallel SECONDO is deployed by Data Servers A Data Server is the minimum execution unit of the system containing a compact SECONDO named Mini SECONDO and its database together with a PSFS node It is possible that one cluster node contains several Data Servers This is especially useful for nodes with multiple hard disks since then the user can set a Data Server on each disk in order to take full advantage of the cluster resources When processing various parallel queries a small amount of data is exchanged among nodes through the HDFS in order to assign tasks to Data
14. as the arguments for the interFunc Both inputs are redistributed based their partition attributes respectively and make queries more flexible In general intermediate data produced by Mini SECONDO databases are shuffled in Parallel SECONDO through PSFS At the same time for the purpose of research Parallel SECONDO also can shuffle inter mediate data only by HDFS This is named HDJ Hadoop Distribute Join since the data are distributed by the Hadoop system This feature is only supported in the hadoopReduce2 operator by setting its optional parameter isHDJ to be true which is false by default More details about HDJ can be found in our technical report about Parallel SECONDO Simple and Efficient Coupling of Hadoop With a Database Engine 6 A Tour in Parallel Secondo In this section a small example is prepared to demonstrate how to use Parallel SECONDO to process queries like building and searching a distributed index or making a distributed hash join query restore database opt from opt open database opt let Con PLZ 16928 let PartFile plz feed spreadl PLZ 6 TRUE let PartRel PartFile hadoopMap consume close database In the above queries the relation plz in database opt is distributed over the cluster It is first distributed by spread into a 6 x 1 PS Matrix with the DLF fiist PartFile as the result Then the PartFile is sent to hadoopMap and its sub files are loaded into slave databases and the
15. e Besides the NS4Master can also be set as true if the user want to visit the existing SECONDO database on the master node Particularly the following five parameters in the parallel configure file should be changed by re placing the localhost with the master node s IP address hdfs 192 168 1 1 49000 hdfs site xml dfs http address 192 168 1 1 50070 hdfs site xml dfs secondary http address 192 168 1 1 50090 mapred site xml mapred job tracker 192 168 1 1 49001 mapred site xml mapred job tracker http address 192 168 1 1 50030 l fs default name core site xml Additionally the transaction feature should also be disabled by uncommenting the line in the file SECONDO_BUILD_DIR bin SecondoConfig ini on the master node RTFlags SMI NoTransactions At last run the ps cluster format script cd SECONDO_BUILD_DIR Algebras cp ParallelSecondoConfig ini S ps cluster format Hadoop clusterManagement ECONDO BUILD DIR bin Activate the Hadoop and HadoopParallel algebras into the file makefile algebras recompile SEC ONDO and distribute Mini SECONDO to all DSs at last cd SS make cd S SECONDO BUILD DIR Algebras Hadoop clusterManagement ps secondo buildMini co ECONDO_BUILD_DIR monn wn The parameter c in the ps secondo buildMini indicates the whole cluster Till now Parallel SECONDO is completely installed Accordingly the user can also easily remove the system with the ps cluster uninstall sc
16. e scripts include at least ps cluster format It is an automatic install script for initializing the Parallel SECONDO environ ment of the cluster including the Hadoop framework ps cluster uninstall It performs the opposite function to the script above removing the Parallel SECONDO from the cluster and cleaning up the environment ps secondo buildMini It extracts and distributes the Mini SECONDO based on the local SEC ONDO system to all Data Servers that are listed in the DS Catalog ps startMonitors The SECONDO monitor is a database server process prepared to accept multiple remote clients visiting the same SECONDO database In Parallel SECONDO all SECONDO monitors have to be started up before processing any parallel queries Considering there may exist several Data Servers in one cluster node this script helps the user to start up monitors on the current node ps start AllMonitors It starts up all Data Servers SECONDO monitors in the cluster ps stopMonitors It shuts down Data Servers SECONDO monitors on the current node ps stop AllMonitors It shuts down all Data Servers SECONDO monitors in the cluster ps startTTY It starts up a SECONDO text terminal interface of one Mini SECONDO on the current node ps startTTYCS It opens a SECONDO Client Server based text terminal interface and connects to a started SECONDO monitors in the cluster ps cluster queryMonitorStatus It checks the status of all SECONDO monitors i
17. eby there are two kinds of flist objects in Parallel SECONDO 1 Distributed Local Objects DLO A DLO divides a large sized SECONDO object to a N x 1 PS Matrix each row is saved in a slave Mini SECONDO database as SECONDO objects called sub objects All sub objects belonging to a same flist have the same name in different slave databases Theoretically DLO flist can wrap all available SECONDO data types 2 Distributed Local Files DLF Data are divided into a R x C PS Matrix and each piece is exported as a disk file in PSFS called sub file Sub files can be exchanged among Data Servers during parallel procedures At present only relations can be exported and kept as sub files hence DLF flist is prepared for relations only Although it is possible to wrap any SECONDO data type with an flist object there are some data too small to be distributed in this way For example a rectangle query window is used by every slave data server during a parallel query but it is not advisable to divide it into smaller pieces which are then distributed over the cluster since the rectangle itself is very small In this case this rectangle can be simply dupli cated to every Data Server during the runtime Apparently not all SECONDO objects can be duplicated since data are delivered as nested lists which require a relatively expensive transforming overhead E g a relation containing one million tuples should not be delivered to every slave along with t
18. hadoopReduce also takes one flist as the input delivering and processing its interFunc by slaves in parallel However the interFunc is processed in the reduce step instead of the map step of its template Hadoop job In the map step the input has to be redistributed based on the partitionAttribute hence the input flist object must wrap a stream of tuples Compared with the hadoopMap operator this operator needs two additional arguments partAttribute and reduceTaskNum Data are first re distributed into reduceTaskNum columns based on the partAttribute in the map step Then each reduce task collects one column data from the re distributed PS Matrix using it as the input for the interFunc Take the third BerlinMOD query as an example which can be processed with the following parallel query let OBACRres003 QueryLicences_Top10_List hadoopMap DLF false feed loopjoin para dataSCcar Licence btree List para dataSCcar List exactmatch Licence LL projectextendstream Licence_LL UTrip units Journey LL extend Box scalerect bbox UTrip CAR WORLD X SCALE CAR WORLD Y SCALE CAR WORLD T SCALE xtendstream Cell cellnumber Box CAR WORLD GRID 11 hadoopReduce Cell O3 Result DLF REDUCE SCALE para OuerylInstants Topl0 Dup List feed II product projectextend Licence Licence LL Instant Instant II Pos val UTrip atinstant Instant_IT filter isdefined Pos collect sort rdup consume
19. he query There fore a new data kind called DELIVERABLE is introduced for Parallel SECONDO and only data types associated with this kind can be used in parallel queries and duplicated to slaves during the runtime All DELIVERABLE data types are listed in Table 2 5 3 Operators Along with the creation of the flist data type several operators are introduced to process distributed objects and parallel queries Briefly these operators are divided into four kinds flow assistant Hadoop and PSFS The PSFS operators are invisible to users exchanging data between SECONDO instances and PSFS therefore they are not specially introduced here 5 3 1 Flow Operators Flow operators connect sequential queries with parallel queries At present two operators named spread and collect are defined for this kind and they can only process DLF flist objects Sub objects in one DLO flist are kept in slave databases and cannot be transferred over the nodes hence they are not supported by the flow operators spread stream tuple T x fileName string x filePath text x dupTimes int x AI x scaleN int x KPAI bool x AJ x scaleM int x KPAJ bool flist stream tuple T spread partitions a SECONDO relation into a PS Matrix distributing pieces into the cluster and returning a DLF flist The relation is first divided to rows in the PS Matrix according to the indispensable partition attribute AJ If another partition attribute AJ i
20. ith the following command cd HOME env grep PARALLEL_SECONDO PARALLEL SECONDO MASTER conf master PARALLEL_SECONDO_CONF conf PARALLEL SECONDO BUILD DIR secondo PARALLEL SECONDO MINIDB NAME msec databases PARALLEL SECONDO MINI NAME msec PARALLEL SECONDO PSFSNAME PSFS PARALLEL SECONDO DBCONFIG SecondoConfig ini PARALLEL SECONDO SLAVES conf slaves PARALLEL SECONDO MAINDS dataServerlf PARALLEL SECONDO dataServerl PARALLEL SECONDO DATASERVER NAMES The third step initializes the DS in the computer but the Mini SECONDO system is not distributed into those Data Servers yet This is because the Hadoop algebra cannot be compiled before setting up the environment of Parallel SECONDO Now both Hadoop and HadoopParallel algebras should be activated and SECONDO should be recompiled The new algebras are activated by adding the following lines to the algebra list file SECONDO_BUILD_DIR makefile algebras ALGEBRA_DIRS Hadoop ALGEBRAS HadoopAlgebra ALGEBRA_DIRS HadoopParallel ALGEBRAS HadoopParallelAlgebra 16 After re compiling the SECONDO system the user can distribute Mini SECONDO to all local DSs with the auxiliary tool ps secondo buildMini cd SSECONDO BUILD DIR make cd SSECONDO BUILD DIR Algebras Hadoop
21. mons with different port numbers In this case the port numbers listed in the second part should also be adjusted The cluster section lists all involved Data Servers Each Data Server is indicated with one line and assigned as the master or a slave by the title The value part contains three elements IP Address Data Server Path and Mini SECONDO Port The second element indicates a disk path where all Data Server components are kept This path is created automatically if it does not exist before but the user should guarantee that he she has been granted the write access to this path Take a cluster configuration for example Master 192 168 0 1 Home dataServerl 11234 Slaves 192 168 0 1 Home dataServerl 11234 Slaves 192 168 0 1 Home dataServer2 12234 Here a small cluster is simulated on one computer with the IP address 192 168 0 1 It contains two Data Servers which are kept in the Home dataServer and the Home dataServer 2 respectively The first Data Server is used as the master and a slave at the same time The master Mini SECONDO can be accessed through the port 11234 The last section options is prepared for some special settings in the cluster At present only one option named NS4Master Normal Secondo For Master database is provided here If its value is set as true then the master node s default SECONDO i e where the SECONDO BUILD DIR points to is set to be the master database 4 Mini Secondo Management In
22. n Ubuntu the following line in the profile file HOME bashrc source SHOME secondorc SHOME secondo should not be set at the end of the file Instead it should be set above the line z SPS1 amp amp return 2 Download Hadoop Go to the official website of Hadoop and download the Hadoop distribution with the version of 0 20 2 The downloaded package should be put into the SECONDO BUILD DIR bin directory without changing the name cd SSECONDO BUILD DIR bin wget http archive apache org dist hadoop core hadoop 0 20 2 hadoop 0 20 2 tar gz 3 A profile file named ParallelSecondoConfig ini is prepared for setting all parameters in Parallel SECONDO Its example file is kept in the clusterManagement folder of the Hadoop Algebra which is basically made for a single computer with Ubuntu However few parameters still need to be set according to the user s computer e The Cluster parameter indicates the DSs in Parallel SECONDO In a single computer it can be set like Master lt IP gt lt DS_Path gt lt Port gt Slaves lt IP gt lt DS_Path gt lt Port gt The IP is the IP address of the current computer The DS_Path indicates the partition for a Data Server for example tmp Note that the user must have the read and write access to the DS Path The Port is also a port number prepared for a DS daemon like 11234 Different DS can be set on the same computer but their DS_Paths and Ports m
23. n contrast if the l parameter is set then the new SECONDO is only distributed to all Data Servers on the current node All essential components that a Mini SECONDO needs are listed in a text file named miniSec_list which is kept together with the script ps secondo buildMini describing files and folders by lines It is possible for the user to change this list according to his own requirement 4 2 Start Up and Turn Off Mini Secondo Monitors During parallel procedures SECONDO databases are accessed through their monitors which should be all started up before processing any queries Considering there are tens or even hundreds of Mini SECONDO systems inside a cluster several utilities are proposed to start and stop these monitors without visiting them one after another The ps startMonitors starts up all Mini SECONDO monitors on the current computer while ps start AllMonitors visits every node and runs the ps startMonitors so as to start up all monitors on the cluster In contrast the ps stopMonitors turns off all Mini SECONDO monitors on the current computer and all monitors on the cluster are shut down by executing the ps stop AllMonitors 4 3 Open Parallel Secondo Interface In Parallel SECONDO every Mini SECONDO can be viewed as a normal SECONDO system and visited independently Normally the user only needs to visit the master database although scripts like ps startTTYCS are capable to access any Mini SECONDO in the cluster O
24. n the cluster 3 Parallel Configuration File All Parallel SECONDO parameters are set in a file named ParallelSec ondoConfigure ini Its example file is also kept in the Hadoop algebra together with the above auxiliary utilities 4 Template Hadoop Job Several generic Hadoop jobs are prepared for the PQC operators Their source files are kept in the Hadoop algebra too and the jobs are generated when the algebra is being compiled 3 Configuring Parallel Secondo All Parallel SECONDO related parameters are set in a file named ParallelSecondoConfig ini Its example file is prepared to deploy the Parallel SECONDO on a single computer and kept in the Hadoop algebra by default After setting all required parameters according to the user s own environment this file should be copied to the SECONDO BUILD DIR bin and then read by the ps cluster format script This file follows the same format as the SecondoConfig ini file mainly divided into three sections hadoop cluster and options The first hadoop section is prepared for setting up the Hadoop framework All parameters listed in this section keep the Hadoop configuration on all involved nodes the same Although it restricts the flexibility of the system it helps new users to set up Parallel SECONDO quickly without learning many details about Hadoop All parameters are listed by lines composed of three elements including the file name title and value Each value is prepared for a parameter wi
25. n two functions d x and d y The d x divides the data into R rows each row should be completely kept on one Data Server It is possible for d z to produce more rows than the cluster scale N and keep multiple rows on one Data Server Afterwards d y further divides every row to C columns As a result a PS Matrix is composed by R x C pieces but not all pieces in the matrix must contain data Normally a PS Matrix is prepared for distributing large sized data over the cluster like relations containing millions of tuples while the division functions are hash algorithms based on one of its attributes 5 2 Data Type In Parallel SECONDO data type flist is especially introduced for representing the PS Matrix It is designed as a wrap structure and able to encapsulate all available SECONDO objects shown in Table 1 After a SECONDO object has been divided into a PS Matrix piece data are distributed and kept in slave Data Servers while only their partition scheme is kept in the meta database as an flist object SPATIAL point line FLIST flist point flist line RELATION rel twple T FLIST flist rel tuple T INDEX rtree T FLIST flist rtree T Table 1 Some Example flist Data Types There are two kinds of methods keeping distributed data on slave data servers The first keeps the partial data in slave Mini SECONDO databases saved as normal SECONDO objects The second exports data to the PSFS nodes as disk files Her
26. ne cluster node may contain several Mini SECONDO databases which are arranged according to the DS Catalog The commands for opening the Parallel SECONDO main text interface are cd SSECONDO_BUILD_DIR Algebras Hadoop clusterManagement ps startMonitors ps startTTYCS s 1 All Mini SECONDO monitors must be started up before running any parallel queries hence the meta database can only be visited through its client server interface Since the master Mini SECONDO is always the first database on the master node we start up this specific interface by setting the argument s with the value 1 Besides it is also possible to use the graphical user interface in Secondo to visit the meta database The user can open the Javagui interface as usual then connect to the indicated database by setting its host IP and port number in the menu Server and Settings 5 Secondo Parallel Query Expression Here we introduce how to write parallel queries in SECONDO executable language The SECONDO exe cutable language describes query plans by composing database objects and operators in certain oder It Figure 2 PS Matrix is of course more complicated than writing SQL queries but far more easier than programming Hadoop programs 5 1 PS Matrix In Parallel SECONDO data is distributed over the cluster as a PS Matrix shown in Figure 2 A Secondo object is divided into pieces based o
27. nuous slave nodes and the default value of dupTimes is 1 All duplicated files are kept in PSFS nodes collect flist stream T x row int x column int stream tuple T The collect operator performs the opposite function to the spread It accepts a DLF kind flist object collects required sub files over the cluster and returns a tuple stream from sub files Both the row and column arguments are optional arguments and their default values are 0 which means the complete row and column If there is only one parameter given then it is viewed as a row number Only non negative integer numbers are accepted as parameters By default this operator reads all sub files denoted in the given DLF flist If the optional parameters are set then it returns part of the PS Matrix For example e collect 1 and collect 1 0 read all sub files in the first row e collect 0 2 reads all sub files in the second column e collect 1 2 reads one piece sub file in the PS Matrix located at the first row and the second column e collect 0 0 and collect read all sub files inside the PS Matrix If the required sub files are located in a remote node then they are copied to the current node before being read 5 32 Assistant Operators All assistant operators cannot be used alone but have to work with the following hadoop operators para ELISE T Se T The flist is designed to wrap all available SECONDO data types and work with various SECONDO
28. p of the profile file HOME bashrc before z SPS1 amp amp return Installation Steps One computer of the cluster must be indicated as the master node and the complete installation is done on this machine only with the following steps 1 Enter the master node download the latest SECONDO distribution which must be newer than 3 3 2 Then extract the SECONDO archive to the master s SECONDO_BUILD_DIR and compile it Af terward download the required Hadoop archive to SECONDO_BUILD_DIR bin tar xzf secondo v33 tar gz cd SSECONDO BUILD DIR make 17 cd SSECONDO BUILD DIR bin wget http archive apache org dist hadoop core hadoop 0 20 2 hadoop 0 20 2 tar gz 2 Prepare the ParallelSecondoConfig ini file according to the environment of the cluster For example the cluster can be set as Master 192 168 1 1 disk1 dataServer1 11234 Slaves 192 168 1 1 diskl dataServerl 11234 Slaves 192 168 1 2 diskl dataServerl 11234 Slaves 192 168 1 1 disk2 dataServer2 14321 Slaves 192 168 1 2 disk2 dataServer2 14321 Here a cluster with two computers and four DSs are described The two computers are set with the IP address of 192 168 1 1 and 192 168 1 2 respectively Since every computer has two disks each is set with two DSs The computer 192 168 1 1 is set to be the master node of the cluster and its first DS with the port of 11234 is set to be the master and the slave DS at the same tim
29. pub gt gt ssh authorized keys Afterward try to ssh to the local computer again this time it may asks to add its current IP address to the known_hosts list like S ssh lt IP gt The authenticity of host can t be established RSA key fingerprint is Are you sure you want to continue connecting yes no This step happens only once when the ssh connection is built at the first time The user can simply confirm the authentication by typing yes If the user prefers to avoid this step the following three lines can be added into the file HOME ssh config Host x StrictHostKeyChecking no UserKnownHostsfile dev null 14 Installation Steps A set of auxiliary tools which practically are bash scripts are provided to help the user to install and use Parallel SECONDO easily These tools are kept in the Hadoop algebra of SECONDO 3 3 2 in a folder named clusterManagement The installation of Parallel SECONDO on a single computer includes the following steps 1 Install SECONDO The installation guide of SECONDO for different platforms can be found on our website and the user can install it as usual Afterward the user can verify the correctness of the installation and compile SECONDO env grep SECONDO SECONDO_CONFIG secondo bin SecondoConfig ini SECONDO_BUILD_DIR secondo SECONDO_JAVA java SECONDO_PLATFORM cd SSECONDO_BUILD_DIR S make Particularly i
30. ript The user can start all SECONDO monitors with the script ps start AllMonitors and close them with ps stop AllMonitors The text interface of the whole Parallel SECONDO is still opened by connecting to the master DS All these steps should and only need to be done on the master node 18 start all sh cd SSECONDO_BUILD_DIR Algebras Hadoop clusterManagement ps start AllMonitors ps cluster gueryMonitorStatus ps startTTYCS s 1 OR OE TEE Sometimes it is guite difficult for the user to build up a physical cluster Regarding this issue a Amazon AMI of Parallel SECONDO is provided With this image the user can guickly set up a virtual cluster on the Amazon Web Services AWS and the Parallel SECONDO is already built inside it hence he she can test the system directly 19
31. s indicated each rowcan be further partitioned into columns Each piece of the PS Matrix is exported as a sub file Both the fileName and the filePath arguments are optional If they are not indicated sub files are then kept in the default PSFS nodes and their names are set by rules The user is allowed to set the file name and path by himself although it may create homonymic sub files with different queries Type Algebra NumOfFlobs PersistencyMode 1 bool StandardAlgebra 0 Memoryblock fix core 2 cellgrid2d TemporalAlgebra 0 Memoryblock fix core 3 cluster TopRelAlgebra 0 Memoryblock fix core 4 duration DateTimeAlgebra 0 Serialize fix core 5 edge GraphAlgebra 0 Memoryblock fix core 6 geoid SpatialAlgebra 0 Memoryblock fix core 7 gpoint NetworkAlgebra 0 Memoryblock fix core 8 ibool TemporalAlgebra 0 Memoryblock fix core 9 iint TemporalAlgebra 0 Memoryblock fix core 10 instant DateTimeAlgebra 0 Serialize fix core 11 int StandardAlgebra 0 Serialize fix core 12 ipoint TemporalAlgebra 0 Memoryblock fix core 13 ireal TemporalAlgebra 0 Memoryblock fix core 14 istring TemporalExtAlgebra 0 Memoryblock fix core 15 point SpatialAlgebra 0 Memoryblock fix core 16 real StandardAlgebra 0 Serialize fix core 17 rect RectangleAlgebra 0 Memoryblock fix core 18 rect3 RectangleAlgebra 0 Memoryblock fix core 19 rectd RectangleAlgebra 0 Memoryblock fix core 20 rect8
32. th the specific title in a particular file fileName title value This section is further divided into four parts The first contains all indispensable parameters prepared for Parallel SECONDO Non advanced users should keep this part unchanged or else the Parallel SEC ONDO may not work correctly The second part sets IP addresses and ports for different Hadoop dae mons They are also indispensable although the user should change their values based on his own cluster like the master node s IP address Besides daemons port numbers can also be changed if the default values have already been taken by some other programs The third part indicates the capability of the cluster and the user should also set them based on his own cluster For example the option mapred site xml mapred tasktracker map tasks maximum limits the number of map tasks running in parallel on one node Usually we set this value by doubling the number of the processor cores contained in the computer so does the other option mapred tasktracker reduce tasks maximum There are also some other parameters like the hdfs site xml dfs replication telling how many times each HDFS block is replicated on the cluster If the Parallel SECONDO is deployed into a cluster composed of hundreds of elastic computers then it is better to set this parameter with an integer more than 1 The last part is prepared for clusters shared by multiple users where each user should set their dae
33. ust be different The version must be 3 3 2 or higher This parameter must be set before continue 15 Set hadoop env sh JAVA_HOME to the location where the JAVA SDK is installed e The user might already have installed SECONDO before and created some private data in the database If so the NS4Master parameter can be set true in order to let Parallel SECONDO visit the existing databases NS4Master true Note here if there is no SECONDO databases created before and the NS4Master parameter is set to be true then the below installation will fail e At last the transaction feature is normally turned off in Parallel SECONDO in order to im prove the efficiency of exchanging data among DSs For this purpose the following RTFlag parameter in the SECONDO configuration file should be uncommented RTFlags SMI NoTransactions The file is named SecondoConfig ini being kept in the SECONDO BUILD DIR bin After setting all required parameters copy the adjusted profile file to ECONDO BUILD DIR bin and start the installation with the auxiliary tool ps cluster format cd SSECONDO_BUILD_DIR Algebras Hadoop clusterManagement cp ParallelSecondoConfig ini SECONDO_BUILD_DIR bin ps cluster format During this installation all Data Servers will be created and the Hadoop is installed Besides the Namenode in Hadoop is formatted at last Close the shell and start a new one Verify the correctness of the initialization w
Download Pdf Manuals
Related Search