Adaptec Snap Server 18000
Contents
1. Table 4 illustrates the resiliency factors tested a description of Table 3 Test Steps Interlink waited for a period of time after the completion of each factor tested in order for the system to normalize 4 Test Cases and Results This section describes the read and write test cases and testing results Interlink used to characterize performance of the Snap Server 18000 with Microsoft SOL Server 2000 Server 2000 Step Description Procedure Comments Open performance monitor and Under Performance Logs and Alerts create a new Once created the same log settings will be used for every test 1 configure counters Log Settings under Counter Logs Save these log settings to be used for all tests Execute Test Script 2 Write to Using Microsoft SQL Server 2000 Enterprise Saved table in the destination database will be called exam 2 appropriate database Manager use DTS to import data to appropriate ples database Start Performance Counter Right mouse click on the saved log settings and Interlink started the perfmon after the test script was executed 3 choose start to ensure no O values would occur in the test data 4 When test script ends stop per Right mouse click on saved log settings and choose Interlink stopped the perfmon log immediately after the test formance monitor stop script ended to ensure no O values would occur in the test data Rename Performance monitor log U2. some with iSCSI initiators and some with iSCSI targets Adaptec has been awarded the Designed for Windows logo on several of their products making them ideal for storing Microsoft SQL Server 2000 data Interlink s testing validates that the Adaptec Snap Server iSCSI device is a viable alternative as a storage solution in Microsoft SQL Server 2000 environments Is it time for you to evaluate iSCSI for your storage needs For those companies that might still be looking to replace older DAS storage or have avoided moving to SAN technology due to high cost and complexity Snap Server iSCSI devices should definitely be considered P N 666766 011 Printed in USA 02 05 3702_1 1 15
3. drive in a RAID 1 configuration Dedicated database drive in a RAID 5 configuration The setup for baseline testing consisted of the creation of 2 RAID volumes on the ILGTESTSS server and installing Microsoft SQL Server 2000 along with the latest service pack SP3a The default C drive was used only to host the operating system and to run the application executables The first volume was created using two internal disks configured in a Raid 1 mirrored set for the transaction logs The second volume was created using eight drives configured in a Raid 5 stripped with parity set for the database files The drives were logically assigned the L drive for the transaction logs and the G drive for the database files Table 5 below lists the base line performance test results 4 1 2 Test 1 Logs amp Database on iSCSI Disks To create this scenario two volumes were created on the Snap Server 18000 One disk set was created with two drives con figured in a RAID 1 mirrored set to hold the transaction logs Table 5 Baseline Performance Test Results INTERLINK A second disk set comprised of six hard drives in a RAID 5 set was created to hold the database Both disk sets had a single volume created on them An iSCSI volume was associated with each respective volume The test Microsoft SQL Serv er 2000 ILGTESTS3 was configured to attach to the two iSCSI volumes using the Microsoft iSCSI initiator component so that they are
4. iSCSI Initiator log onto the appropriate iSCSI volumes 5 Format and assign a drive letters to the mounted volumes 6 Create database using the new drive letters 7 Add new databases to existing backup jobs 5 3 6 iSCSI Initiator Startup Procedures iSCSI volumes are mounted later in the Windows boot process Without special steps being taken Microsoft SQL Server 2000 can start before the iSCSI volumes are mounted and databases that reside on the Snap Server 18000 will be in sus pect mode within Microsoft SQL Server 2000 To ensure that Microsoft SQL Server 2000 does not start before all iSCSI volumes are mounted follow these steps 1 When logging onto the appropriate iSCSI volumes in the Microsoft iSCSI Initiator Properties program choose Automatically restore this connection when the system reboots thereby making this a persistent target 2 Set the MSSQLServer service dependent on the Microsoft iSCSI Initiator service 3 Set the Bind Volumes option in the Microsoft iSCSI Initiator Properties on the Initiator Settings tab 5 4 iSCSI Protocol Microsoft SQL Server 2000 requires that databases be creat ed on disk volumes using drive letter assignments in Disk Management iSCSI allows Snap Server 18000 iSCSI format ted volumes to appear to the operating system as a local disk volume iSCSI is the implementation of SCSI commands wrapped in TCP IP packets TCP handles flow control and ensures
5. network interface card NIC in iSCSI scenarios When complete repair the network cable by plugging the cable back in to the Snap Server 18000 system head unit This will restore the disk subsystems to pre test condition 5 Power off Snap Server 18000 and baseline external SCSI This will fracture the communications between the Snap Server 18000 system and the enclosure unexpectedly during database testing external SCSI storage with the host system and affecting disk operations SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION 4 1 Read amp Write Tests Interlink used the built in Performance Monitor utility in Windows Server 2003 to monitor log and average various sys tem performance counters including CPU parameters physical disk performance and network performance 4 1 1 Baseline Test Database amp Logs on Internal Direct Attached Drive Prior to Adaptec scenario testing a baseline configuration matching a typical customer scenario was tested in order to provide a known condition against which to measure The configuration used to represent a typical customer s Microsoft SOL Server 2000 is as follows Compaq Proliant DL 380 dual processor Pentium III Serv er class 1GB RAM Integrated non accelerated RAID card for internet disk hardware accelerated RAID card with battery backed cache for external expansion chassis Dedicated OS Application drive Dedicated log
6. presented as local disk Once the volumes are configured on the ILGTESTS3 server the transaction log drive was assigned the H drive letter and the data drive was assigned the I drive letter Interlink then cre ated the three databases 1GB 10GB and 100GB using these drive letters for this test The Test 1 configuration attempted to simulate the perform ance of the Snap Server 18000 device when it was configured similar to the local server The transaction log files and data base files are all stored on the Snap Server 18000 s storage system in comparison to the transaction log files and databas es being all local to the server on SCSI The disks within the Snap Server 18000 are running at 7200 RPM they are not as fast as the 10 000 RPM Wide Ultra SCSI disks Table 6 below lists the Test 1 configuration performance test results Using our top 3 performance indicators Figure 2 illustrates the percent difference for all database sizes in the Test 1 configura tion compared to the baseline configuration Test 1 Logs amp Database on iSCSI Disks im 100GB Test 1 W 10GB Test 1 im 1GB Test 1 35 30 25 20 15 10 5 0 Delta to Baseline Figure 2 Test 1 Configuration Performance Test Results oie Se SS Ten pr pessum quare 1GB Base 487446 422 8179131 174 22248813 04 0 61 28 58 54 39 86 22 10GB Base 290776 324 7526451 27 22920794 11 0 61 1 53 60 49 89 38 100GB Base 320404 18 799105
7. the database 4 1 4 Test Results Comparison The six charts shown in Figure 4 on the next two pages illus trates the 1GB 10GB and 100GB performance characteristics of the two Snap Server 18000 test configurations to the base line configuration In order to understand disk throughput Interlink captured database write bytes per second database read bytes per second and log write bytes per second Interlink discovered two similarities from the performance test results 1 The baseline configuration tests were faster than the two Snap Server 18000 test configurations and 2 The two Snap Server 18000 test configurations perform ance database read results were basically equal with improvements in the Test 2 scenario for database writes Avg Disk Write Processor Time Processor Time Queue Length Read Write Avg Disk Read Queue Length 10GB Test 2 313329 314 7588861 439 17787942 92 62 34 87 36 293437 398 7385352 749 12230381 73 100GB Test 2 SQL Server 2000 ISCSI Snap SERvER 18000 CHARACTERIZATION N I f n N K Figure 4 1GB 10GB and 100GB Performance Test Results 1GB Read amp Write Performance Comparison umm DB Write Bytes Sec E DB Read Bytes Sec LG Write Bytes Sec 0 000 1GB Base 1GB Test 1 1GB Test 2 Test Configuration 1GB DB Read Write amp LG Write Test Comparison m Test 2 Database on iSCSI Disks amp Logs on hternal Direct Attached
8. transaction logs for the Test 1 configuration One RAID 5 set was created using the remaining six 250GB disks to hold the databases for Tests 1 and 2 configurations For all tests Microsoft best practices were followed by separat ing transaction logs sequential data access and database files random data access onto different physical drives Table 1 lists the logical drive letter allocation the RAID level the num ber of drives and the purpose of each drive Figure 1 illustrates the Interlink test lab hardware configuration Public Network SMC 1000 MB Switch Compaq Proliant DL 380 Windows 2003 Dual 733 MHz EECOP PESE eit f 1 GB RAM d Integrated Array i da a TOC SNAP900148 x Linux Kernel Smart Array 3200 Dual 2 4 GHz Intel Xeon processors U2 SCSI Expansion Chassis 2GB Double Data Rate DDR system memory 18 2 GB 10Kx12 512 MB of battery backed non volatile memory 100 Mb NIC 250 GB 7K drives x8 1000 MbSMCEZNIC Dual Gigabit Ethernetports Figure 1 Iest Lab Environment Table 2 lists the servers and devices used within the test lab environment platform OS software TCP IP address and a description of the function performed 3 3 Read and Write Test Approach Using the test scripts in section 3 3 1 Interlink benchmarked e g baseline configuration the read and write performance characteristics of a 1GB 10GB and 100GB database placed on Purpose Number of Drives Test Scenari
9. 9 35 15543701 09 1 52 37 83 60 43 84 12 Table 6 Test 1 Performance Test Results eres ne a 1GB Test 1 418537 963 7723526 997 15717719 11 0 31 8 64 61 56 88 67 10GB Test 1 273422 348 6635551 485 17650823 83 0 51 7 66 61 59 86 73 100GB Test 1 279209 936 7087308 909 11703422 65 0 28 16 35 59 28 84 98 SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION INTERLINK Test 2 Database on iSCSI Disks amp Logs on Internal Direct Attached Drive Delta to Baseline Figure 3 Test 2 Configuration Performance Test Results 4 1 3 Test 2 Database on iSCSI Disks amp Logs on Internal Direct Attached Drive To create this scenario Interlink placed the transaction logs for these databases on the locally attached L drive and placed the database files on the I drive This design splits the data base between locally attached storage for the transaction logs and the Snap Server for the database files Interlink wanted to test this scenario to see if this configuration would offer any performance benefits by having transaction logs on local disk The Test 2 configuration attempted to present a more normal ized yet more optimized configuration by placing the transac tion logs on the fastest drives within the server and the data base on the most expandable system Snap Server 18000 Interlink initially assumed that this configuration would per form slightly bett
10. Drive B Test 1 Logs amp Database on iSCSI Disks 35 30 25 20 15 10 5 0 Delta to Baseline 10GB Read amp Write Performance Comparison mum DB Write Bytes Sec Ez DB Read Bytes Sec LG Write Bytes Sec 0 000 10GB Base 10GB Test 1 10GB Test 2 Test Configuration INTERLINK SQL Server 2000 I SCSI Snap SERVER 18000 CHARACTERIZATION 10GB DB Read Write amp LG Write Test Comparison m Test 2 Database on iSCSI Disks amp Logs on Internal Direct Attached Drive HB Test 1 Logs amp Database on iSCSI Disks 25 20 1596 1096 596 096 596 10 Delta to Baseline 100GB Read amp Write Performance Comparison 18000000 000 16000000 000 14000000 000 12000000 000 10000000 006 mmm DB Write Bytes Sec DB Read Bytes Sec 8000000 000 LG Write Bytes Sec 6000000 000 4000000 000 2000000 000 0 000 100GB Base 100GB Test 1 100GB Test 2 Test Configuration 100GB DB Read Write amp LG Write Test Comparison m Test 2 Database on iSCSI Disks amp Logs on hternal Direct Attached Drive n Test 1 Logs amp Database on iSCSI Disks 30 25 20 15 10 5 0 Delta to Baseline 11 SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION Interlink did expect the baseline configuration to outperform the two Snap Server 18000 test configurations since the SCSI interface and interconnect are faster than 1000Base T Ether net and the SCSI Ultra Wide drives are fast
11. EN edd Sn 5 5 Microsoft Windows 1SCSI Initiator Service ssscsisessisesssessssescossisscannssssccasssssosseassensisecsensstsossassseuasasssnesiduasdazaiedaies 15 SEESOGNISNENI GM ie e a N AN EE AN GNE E A GRN 15 SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION Introduction Adaptec is the worldwide volume leader in network attached storage NAS products offering its customers the successful Snap Server family of Network Attached Storage servers These NAS solutions combine the most reliable and cost effective technologies with industry leading ease of integration Adaptec is actively developing new features and capabilities that extend the traditional NAS appliances into an iSCSI target that sup ports both block and file needs By combining these new iSCSI capabilities with industry leading ease of integration Adaptec strongly believes there are significant opportunities to target applications within the enterprise that require block level serv ices In order to target and sell to these enterprise customers Adaptec has positioned its storage appliance in terms of an overall solution The introduction of low cost high performance storage devices provides new opportunities for enterprise IT to use storage in innovative ways that were previously cost prohibi tive or not technically feasible One such opportunity is to leverage iSCSI devices for use in database applications Unlike previous NAS device
12. SQL Server 2000 iSCSI Snap Server 18000 Characterization Prepared for SNAP Appliance by Interlink Group INC February 16 2005 Interlink Group LLC 98 Inverness Drive East Suite 150 Englewood CO 80112 Phone 303 542 7100 Toll Free 888 533 1307 http www interlinkgroup com SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION N I f R N K Contents T INTRODUCTION ennn e EE EEEE ENE EEA EE ET E 2 1 1 Document Overview 3 IAE luco iii M 3 2 SOLUTION OVERVIEW eM 3 2 1 Statement of the Problemi estet tree tege inrer etie race pipa aveo da eeu bn ipe run aso doge 3 2 2 Sol tion DesCDIDEBOD Leer cccauteysnsdtseds reges tie a e decente ali rure eatis a EE eere es DERE 3 2 3 Approach By TES TING METHODOLOGY si asstessissvcaseneconteastabestevenssstsanisseceveseetcdenetohuusbneosetesutaabenskevysthsbeasantusbsteasdhedeeasusnsteducsubateend 4 31 Tool Sele E 4 3 2 Eeviuciaor tp 5 3 3 Readand Write Test Approach scaena D RR ORG O aAA bra re ob E ORE RSS 5 3 3 1 Test Scripts T 3 3 2 Test PLOCCAUTES eene Rete red Tete od Caetera cree eee ERIS Pere torre eet tete tee ERE HH IRE 6 Bid Resiliency Test Approach ice eec iii eive eebecon ier eie ve Yee Det eap ee es deg g tege estes eee eee Lee eben 7 A MINIZdUIe eI NER RR M S 7 4 FEST CASESAND RESULTS De eS Haee xe rta teria eee eei ipu ie
13. ant to migrate to a con solidated storage environment to overcome the inflexibility and maintenance issues of DAS SAN was the answer While SAN s are the most redundant and the fastest storage available they are also the most costly and the most complicated to implement This was all before iSCSI however iSCSI based NAS bring to market a third option To imple ment a Microsoft SOL Server 2000 with a NAS using iSCSI is very easy flexible inexpensive and allows the organization to move to a consolidated storage approach The downside to implementing iSCSI is performance All data packets have to travel over the network which for most organizations is Ether net Further improvements to Ethernet in the future will help However the performance numbers from the tests were good as compared to DAS difference drive throughput on a 3 5 mil lion row select between the DAS and a iSCSI solution was 3 23MB s and may prove suitable for many organizations This is especially true for organizations that are trying to reduce TCO in their storage strategy or cannot afford the large upfront and ongoing maintenance expense of a SAN solution There has been a lot of energy and excitement about iSCSI and it appears that iSCSI is finally coming to fruition after a few slow years of development Microsoft released its software iSCSI initiator in June 2003 A few months later Microsoft awarded its iSCSI Designed for Windows logo to several ven dors
14. eing run add as much memory as desired and be sure to enable AWE Address Windowing Extensions in Microsoft SQL Server 2000 More information about AWE can be found in Microsoft SOL Server 2000 Books Online If there is more than 4GB of RAM consider upgrading to Windows Server 2003 Enterprise Edition as it supports up to 32GB of RAM while Windows 2003 Server only supports up to 4GB of RAM Also when there is more than 4GB of RAM adding Physical Address Extension PAE to the boot ini is required to see the additional RAM All processors based on the IA 32 architecture that begin with the Intel Pentium Pro support 36 bit physical addressing mode known as PAE PAE allows up to 8GB of physical memory on Windows 2000 Advanced Server and up to 32GB of physical memory on Windows 2000 Data center Server and Windows 2003 Server Enterprise Edition 5 3 5 Installation and Volume Creation Procedures Detailed installation and configuration instructions are beyond the scope of this document High level installation and config uration steps are provided to give the reader a general idea of how the solution is deployed The following steps are required to install the solution 1 Install and configure the Snap Server 18000 2 Create RAID sets volumes and iSCSI volumes on the Snap INTERLINK Server 18000 3 Install iSCSI Initiator on Microsoft SQL Server 2000 and configure the Snap Server 18000 as iSCSI target 4 Using the
15. er than Test 1 but still not equal to the base line because of the placement of the transaction logs on local storage The test data does support the theory showing faster log write bytes sec whereas the database read bytes sec were basically equal Table 7 Below lists the Test 2 configuration performance test results Again using our top three performance indicators Figure 3 illustrates the percent difference for all database sizes in the Test 2 configuration compared to the baseline configuration In a Microsoft SQL Server 2000 database a transaction Table 7 Test 2 Performance Test Results DB Write Bytes Sec 1152401 874 DB Read Bytes Sec 18867012 02 Test Performed LG Write Bytes Sec 486914 464 1GB Test 2 i 100GB Test 2 m 10GB Test 2 im 1GB Test 2 must first be successfully written to the log before it will be committed to the database It does this for performance rea sons Writes to a transaction log are sequential and thus very fast If writes were to pass straight to the database first they would be very slow as they would be contending with random reads already occurring on the database In this particular sce nario all writes to the transaction log must first pass through the hardware RAID card to be written to the transaction log then down through the stack to the iSCSI redirector and final ly to the RAID card within the Snap Server 18000 to be writ ten finally to
16. er than the ATA drives in the Snap Server However some Test 2 performance metrics were very close and some slightly better than the base line tests Interlink also expected that Test 2 would outperform Test 1 The performance data shows that in fact Test 2 did outperform Test 1 in regards to log write bytes per second whereas the database read operations were basically equal Because of the test results the optimum choice would be keeping transaction logs on the fastest disks available while moving the larger and growing database data files to the more flexible and more inex pensive iSCSI storage Interlink did expect the network connectivity between the server and the Snap Server 18000 to be a bottleneck of this system One performance counter Network Output Queue Length was observed on the NIC This counter measured data being queued on the NIC that needed to be sent over the net work and was present during the database write operations This does indicate that network bandwidth and latencies were indeed a bottleneck of this design The speed difference is definitely noticeable but when you fac tor the price difference between the two storage options as well as the ability to grow the Snap Server 18000 very easily among multiple servers it is easy to justify the investment in the Snap Server 18000 Table 8 Resiliency Test Results Snap Appliance INTERLINK 4 2 Resiliency Tests The results of the resiliency t
17. erver 2000 has marked the database as suspect the hardware must be repaired and powered on Interlink had to restart the server as the logical drive letter would not re appear The Compaq Array controller initiated a RAID re sync process on all RAID disks in the system By rebooting Microsoft SQL Server 2000 was thus restarted which cleared the database s suspect status No database corruption was seen For the Snap Server 18000 the DTS Write operations halted and Microsoft SQL Server 2000 marked the database as suspect Windows Explorer does not list the Snap Appliance iSCSI mapped drives Now that Microsoft SQL Server 2000 has marked the database as suspect the hardware must be repaired and pow ered on The logical drive letters re appeared without a system reboot The Snap Appliance initiated a RAID re sync process on all RAID disks in the system Microsoft SQL Server 2000 had to be Restarted to clear the database s status as Suspect No database corruption was seen 12 SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION assigned a drive letter Create databases on this new drive letter as if the storage was connected locally The design below illus trates a single Microsoft SQL Server 2000 design with the Snap Server 18000 on a private network External SCSI Public Chassis Network SQL Server Private 1000Mbp Network Laptops Snap 18000 iSCSI Appliance F
18. esting are summarized in this sec tion The results are compared to the baseline and expressed as a Pass or Fail and the description 4 2 1 General Comments Adaptec has taken many great steps to increase the reliability and usability of their Enterprise products The Snap Server 18000 boasts dual processors and dual power supplies both of which are improvements over its predecessor The Snap Serv er 18000 has a very functional internal layout with four hot swappable CPU fans and five hot swappable chassis fans One great feature that is new to Adaptec products is the addition of a LCD screen on the Snap Server 18000 The screen displays the boot up procedures during reboots and displays the server name and IP address when the server is operational 5 Recommended Architecture This section describes the conceptual design system require ments and configuration details Interlink recommends when integrating the Snap Server 18000 with Microsoft SQL Server 2000 to extend the life of an existing database server 5 1 Conceptual Design The flexible inexpensive and simplistic nature of the Snap Server 18000 allows for a quick and easy deployment of a Network Attached Storage NAS combined with the Internet SCSI iSCSI protocol to allow volumes from the NAS device to be mounted to the Microsoft SQL Server 2000 server s as a local volume Once mounted as a local volume it can be Test Baseline Test 1 C
19. how Interlink used these tools to characterize the Snap Server 18000 with Microsoft SQL Server 2000 3 2 Lab Configuration Interlink built the test server ILGTEST3 with Windows Server 2003 Enterprise Edition in a workgroup configuration Microsoft SQL Server 2000 Enterprise Edition was installed and configured using defaults for the Microsoft SOL Server 2000 application file locations For the baseline test Interlink configured two internal drives in RAID 1 for the OS and Microsoft SQL Server 2000 installation The other two inter nal drives were configured in RAID 1 for the transaction logs of the baseline and Test 2 configuration tests Interlink configured the remaining eight external expansion chassis drives in RAID 5 for the database files for the baseline configuration tests Then Interlink installed the Snap Server 18000 SNAP900148 in the test lab and configured it for the same workgroup as test server The Snap Server 18000 was shipped with eight 250GB disks in a default RAID 5 configuration Interlink delet Table 1 Logical Drive Letter Allocation Logical Drive Letter Physical Location RAID Level Internal SCSI External SCSI Chassis Snap Server iSCSI Attached Snap Server iSCSI Attached Table 2 Lab Servers and Devices mus meme dd mus me dS INTERLINK ed this configuration and built two new RAID sets One RAID 1 set was created on the first two disks for the
20. ice destination Finally the network layer typically Ethernet adds information and sends the packet across the physical network to the target storage device 5 5 Microsoft Windows iSCSI Initiator Service The initiator service enables the host computer system to dis cover target storage devices on the storage area network and to determine whether or not it has access to those devices The iSCSI Naming Services iSNS client the code required to interact with the iSNS server is built directly into the initia tor service enabling the initiator service to maintain a list of targets reported via the iSNS server as changes are made to iSCSI volumes on the Snap Server 18000 6 Conclusion Initially Network Attached Storage NAS were used for file based storage that could be shared with Windows clients Unix clients and Apple Macintosh clients easily and cost effectively At that time there were two popular storage choices for Microsoft SQL Server 2000 database storage 1 Direct Attached Storage DAS using SCSI based server integrated RAID array controllers or PCI RAID array controllers with SCSI based storage enclosures or 2 Storage Area Network SAN using Fibre Channel To implement DAS for Microsoft SQL Server 2000 storage is relatively easy moder ately priced performs well but is very inflexible and can be a maintenance challenge when there are many servers each with INTERLINK their own DAS For customers who w
21. igure 5 Conceptual Design Note Additional servers can be implemented by connecting the server to the public network and also to the private network to use storage provided by the Adaptec iSCSI Appliance This allows storage consolidation onto the Snap Server 18000 for other servers while retaining the dedicated private network 5 2 System Requirements The Microsoft SQL Server 2000 configuration for this Con ceptual Design requires Microsoft Windows 2003 Server or Microsoft Windows 2000 Service Pack 4 with the Microsoft iSCSI Initiator installed Version 1 06 of the Microsoft iSCSI Initiator is the recommended level to run for iSCSI connectivity Microsoft SOL Server 2000 A 1000 switched network segment to connect the server to the Snap Appliance solely for the iSCSI protocol Snap Server 18000 with iSCSI support currently the GuardianOS v3 2 019 is the minimum recommended level of the Operating System OS to run for iSCSI con nectivity 5 3 Configuration Details The following section outlines several recommendations and high level implementation procedures to help make imple mentation an iSCSI solution easy and a success 5 3 1 Dedicate Storage Backbone Figure 5 illustrates a simple single server architecture using a dedicated network switch for connectivity between the server and the Snap Server 18000 While a NAS device can be implemented on a public network connecting a Microsoft SQL Ser
22. influence the performance characteristics Interlink used the following query to test and measure read performance Select c_unique c_udist from examples This query returned all rows for the two columns specified Interlink chose columns of varying data types to get a good mixture of data select types 2 Test Script 2 Write Insert Performance Using DTS Interlink copied in bulk a 3 5 million row table 35 million row table and 350 million row table to each test sce nario to test drive throughput to the Snap Server 18000 Interlink decided to use this method to test write performance under heavy conditions to understand drive performance 3 3 2 Test Procedure Interlink ran each test in a very precise and consistent manner to ensure accuracy and comparable results across each test size and test configuration Table 3 lists the steps performed for each test configuration SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION 3 4 Resiliency Test Approach In order to attain consistent results for comparison Interlink introduced a uniform set of faults to the baseline configuration and the Test 1 configuration using the Snap Server 18000 INTERLINK each factor and the order in which they were introduced Prior to executing each fault test on each of the test configura tions Interlink started a 1GB DTS write operation and then a customized performance monitor to gather metrics 3 4 1 Test Procedure
23. isk Due to the large numbers of variables that could impact performance Interlink and Adaptec make no guarantee or claim that similar performance would be achieved in a pro duction environment As they say in the car commercials your actual mileage may vary 3 Testing Methodology This section describes the lab configuration read and write testing approach and the resiliency testing approach Interlink used to characterize performance of the Snap Server 18000 with Microsoft SQL Server 2000 3 1 Tool Selection For benchmarking Microsoft SQL performance Interlink felt that it was important to use tools that actually stressed and tested a live Microsoft SQL Server 2000 environment Microsoft has developed some testing tools available that pro vide synthetic testing that simulates Microsoft SQL Server 2000 T O without actually using the Microsoft SQL Server 2000 application Similarly benchmarks can be found using the IOMETER tool that was originally developed by Intel Although this tool provides good generic I O comparative measurements it does not exercise Microsoft SQL Server 2000 to obtain its data and that data must be extrapolated to predict Microsoft SQL Server 2000 performance character istics Interlink felt that using a tool that did not perform actu al Microsoft SQL Server 2000 operations invalidated the testing After evaluating Microsoft provided tools like SQLIO and SQLIOStress
24. l access iSCSI also elimi nates having to resort to expensive proprietary recovery mechanism and allows the use of third party solutions Existing Infrastructure iSCSI architecture leverages exist ing network infrastructure technology that is well under stood in most organizations and does not require invest ment in alternative storage infrastructures such as Fibre Channel 2 3 Approach As stated above the difficulty in characterizing performance for Microsoft SQL Server 2000 is the myriad of ways in which it can be deployed It would be a daunting task to state definitively that a storage device would perform in a certain way in a unique scenario However it is reasonable to perform specific and defined testing scenarios that would give a poten tial user of an iSCSI storage device a general idea of how that device would perform over direct attached storage DAS At a very generic level most applications of Microsoft SQL Serv er 2000 can be placed into three very broad categories Simple Data Model In this category Microsoft SOL Serv er 2000 is used as a simple repository for data High per formance is not usually essential as the transaction count is relatively low Reads and writes to databases tend to be more sequential in nature Examples of applications that would use SQL in this manner might be SharePoint Portal Server or Microsoft Operations Manager Transactional This category is more perfor
25. lacing the database transaction log files on the fastest disks possible preferably these would be dedicated internal disks in the server and the database data files on the Snap Server 18000 Servers today come with integrated Ultra320 SCSI array controllers with support for 10 000 and 15 000 RPM Ultra320 SCSI drives Purchasing two of these drives in the server and dedicating them to the database transaction log files or keeping the transaction log files where they are currently in an existing server will help ensure a good transition to storing database data files on iSCSI attached Adaptec storage solutions The Snap Server 18000 uses Serial ATA SATA drives running at 7 200 RPM while transferring data to the SQL server over the 10 100 1000 Ethernet network 5 3 4 Optimize Cache Knowing that this solution will introduce a slower disk system extra thought and preparation should be done to optimize caching By implementing this solution a real cost savings is realized and some of these savings can be spent on some more Random Access Memory RAM for the Microsoft SOL Serv er 2000 If the Standard version of Microsoft SOL Server 2000 is being run increase the memory to at least 3GB This will provide Microsoft SOL Server 2000 with 2GB of mem ory with 1GB left for the OS Be sure to configure Microsoft SQL Server 2000 to use the maximum amount of memory If the Enterprise version of Microsoft SOL Server 2000 is b
26. mance depend ent in that a transactional application makes frequent reads and writes Data reads and writes are highly random rather than sequential Usually transactional applications require high availability robust I O performance and rapid recov erability On line business and e Business applications would utilize this model OLAP Microsoft SQL Server 2000 is increasingly being used for Online Analytical Processing which primarily involves aggregating large amounts of diverse data OLAP can involve millions of data items with complex relation ships NAS devices lend themselves to this model in that a point in time snapshot of a production database can be made and analysis run against that near time data with out impacting performance on the production systems INTERLINK Interlink s choice of testing tools reflect a model that encom passes the first two models as they are probably more repre sentative to typical use scenarios where an iSCSI Snap Appli ance device would be deployed Interlink test scenarios includ ed database sizes of 1 10 and 100 gigabytes in various config urations Resiliency of configurations was also tested to simu late component failures and how the Snap Server would han dle the failures It is important to note that Interlink s testing of the Snap Serv er exercised a very limited scenario so that performance char acteristics could be generally identified in comparison to local SCSI d
27. nce To better understand read performance Interlink focused on disk throughput Interlink developed a 1GB database 3 5 mil lion row table a 10GB database 35 million row table and a 100GB database 350 million row table Interlink leveraged the Microsoft SQL Server 2000 Resource Kit DBGen tool to generate random data for the three databases Interlink cre ated a table using a script provided with DBGen Create table examples c_uniqueint c_zdist int c_edist char 255 c_udist varchar 80 c_ndist float c_domain varchar 40 Go This script created a table that is six columns wide with vary ing data types Next Interlink launched the DBGen applica tion and specified the server name the database name the login name and password DBGen requires that the number of rows to create be specified prior to randomly generating data in each of the six columns For example Interlink speci fied that DBGen create 3 5 million rows for the 1GB database With the data randomly created for the three databases the first test used a basic SQL select statement to test disk through put as read operations for the test databases Since the data bases did not have any indexes created on the columns all read operations were complete table scans and provided a good measure of disk throughput Interlink also stopped and start ed Microsoft SQL Server 2000 before each select statement to ensure that SQL caching did not
28. o Baseline Test 2 Baseline Tests 1 Test 2 Server Name Platform OS Software TCP IP Address Description Function Compaq DL380 10 202 2 5 w s 202 2 Workgroup Server ILGTESTS3 Windows Server 2003 Enterprise Edition MI Microsoft SQL Server 2000 Enterprise Edition SP3a 10 200 2 5 Microsoft SQL Server 2000 Database Server Snap Server 18000 device SNAP900148 GuardianOS v3 2 019 10 200 2 4 NAS device a f 192 168 1 2 SMC gigabit 5 port switch management port 10 100 1000 Ethernet Network Switch Note The Snap Server 18000 device only supports RAID levels 0 1 and 5 INTERLINK SQL Server 2000 ISCSI Snap Server 18000 CHARACTERIZATION the L drive transaction log and G drive database run ning on locally attached SCSI storage and compare those num bers to two separate test configurations leveraging the Snap Server 18000 The two specific Snap Server 18000 configu rations included 1 Test 1 Configuration Logs amp Database on iSCSI Disks Using the test scripts in section 3 2 Interlink tested read and write performance characteristics of a 1GB 10GB and 100GB database placed on the H drive transaction log and I drive database 2 Test 2 Configuration Database on iSCSI Disks amp Logs on Internal Direct Attached Drive Using the test scripts in section 3 2 Interlink tested read and write performance characteristics
29. of a 1GB 10GB and 100GB database placed on the L drive transaction log and I drive database Interlink then used the Database Generator sample DBGen available in the ToolsAndSamples DBGen folder on the Microsoft SQL Server 2000 Resource Kit CD to generate 1GB 10GB and 100GB databases for read and write tests The Database Generator is a sample ODBC based tool designed to populate tables in a database with a set of sample data DBGen allows you to choose the type of data distribution you prefer from a list of available distributions The source code for the Database Generator sample is included as a compressed file named Dbgen zip Database Generator requires the following Microsoft SOL Server 2000 A file compression program that supports ZIP compressed files in order to access the sample source code Microsoft Visual Basic 6 0 to access the source code for Dbgen exe Microsoft Visual C 6 0 to access the source code for Dbgencom exe Microsoft SQL Server 2000 security is required to log on when you use the tool Microsoft Windows NT Integrated security is not supported 3 3 1 Test Scripts Interlink created and ran the exact same read and write test scripts against the nine databases to compare the baseline con figuration with each of the two test configurations leveraging the Snap Server 18000 The two specific test scripts included 1 Test Script 1 Read Select Performa
30. onfiguration Comments T Pass Pass For Test 1 configuration when a drive was removed the Snap Server 18000 operated as expected The array was degraded and the SQL operation continued When a drive was re inserted and a repair process was initi ated manually to synchronize the array back to a healthy state the SQL operation continued and the repair process seemed to wait until drive 1 0 operations were limited before beginning the repair process 2 Pass Pass For Test 1 configuration when a drive was removed the Snap Server 18000 operated as expected The array was degraded and the SQL operation continued When a drive was re inserted and a repair process was initi ated manually to synchronize the array back to a healthy state the SQL operation continued and the repair process seemed to wait until drive 1 0 operations were limited before beginning the repair process 3 N A Pass For Test 1 configuration the SQL Operation continued during the degraded network connection state 4 N A Pass For Test 1 configuration the rebuild process continued while the network was in a degraded state There were no write operations occurring during this time from the failures on Test 1 configuration 5 Pass Pass Catastrophic Failure for both configurations is expected For baseline the DTS Write operations halted and Microsoft SQL Server 2000 marked the database as suspect Windows Explorer does not list the SCSI mapped drives Now that Microsoft SQL S
31. re e Cu Pe HR via 7 AlN Read amp aM ai Ac 8 4 1 1 Baseline Test Database amp Logs on Internal Direct Attached Drive ssssseeeee 8 4 1 2 Test 1 Logs amp Database on Mirrored iSCSI Disks essent tenttenttntnntnnen 8 4 1 3 Test 2 Database on iSCSI Disks amp Logs on Internal Direct Attached Drive sss 9 4 1 4 Test Results Comparison iin ctor n e etr FUR ria n iiis DRE REGE RUE ispbieed aazaacabianes 9 ZEIT asasscs sevens oczsssyivashvascasaesssvasasusoscusevsssacosssesasussdavsusetshidecscustensedsns E ESTA 12 42 1 General Comm Ettts cR 12 5 RECOMMENDED ARCHITECTURE ics sssistivesstessedeisdetissdeteiccdetadetadedebsssasedeysastedassuvesacosdded odetedsaustosvdsesttscedeetaiadindede 12 5 1 Conceptual D Slgni csssssss cissasceceessccsseuvseguseseswecsscoudsassedesnters env cdasntasesgeescnssesd cased spe vansvesbveaycasebsceasetusseestvscoveess 12 5 2 System Requurements n itte et iO ER ERE R ER R R RSA RIAR A 13 5 9 Configuration Details 5er teni epe Ud etri et Pe a p RAE OEA 13 5 3 1 Dedicate Storage Backbone sil 3 3 2 Install iSCSI TOE Gated T 13 5 3 Database Architecture uae netten NETE TIEA EAEE 14 b 3 4 Optimize Cache ci eene ritate fete d tede Mete RT Ee REE EEE A PREMO OEGE 14 5 3 5 Installation and Volume Creation Procedures eese tenente tententtntennententennenni 14 5 3 6 iSCSI Initiator Startup Procedures 5 4
32. reliable transmission of data while IP is responsible for the routing of the packets to the destination device iSCSI then contains the SCSI commands such as the destination file target and the actual data The iSCSI protocol provides an end to end pro tocol for transporting commands and block level data down through the host initiator layers and up through the stack lay ers of the target storage devices This communication is fully bidirectional as shown in Figure 6 where the arrows indicate the communication path between the initiator and the target by means of the network 14 SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION Initiator Target Application Logical Unit SCSI SCSI iSCSI protocol iSCSIProtocol TCP IP TCP IP Ethernet Ethernet Network Figure 6 iSCSI Protocol Stack Layers The initiator usually a server makes the application requests These are converted by the SCSI class driver to SCSI com mands which are transported in command description blocks CDBs At the iSCSI protocol layer the SCSI CDBs under control of the iSCSI device driver are packaged in a protocol data unit PDU which now carries additional information including the Logical Unit Number LUN of the destination device The PDU is passed on to TCP IP TCP encapsulates the PDU and passes it to IP which adds the routing address of the final dev
33. riefly describes the technical problem that the solution addresses as well as a general overview of how Inter link approached the generalized performance testing of the Adaptec Snap Server iSCSI device in a Microsoft SQL Serv er 2000 environment 2 1 Statement of the Problem If one were to ask the question is this storage device suitable for use as Microsoft SQL Server 2000 primary storage the answer would have to be it depends There are infinite ways that Microsoft SQL Server 2000 can be used to address common business problems today Each usage case will have its own associated performance requirements that might embrace or preclude the use of one storage method over another Enterprises considering the use of new lower cost network attached storage devices with the iSCSI protocol with Microsoft SOL Server 2000 need to be assured of the per formance characteristics in this database environment When evaluating potential storage solutions for a Microsoft SOL Server 2000 environment the following attributes should be taken into consideration Performance Does the solution meet the performance requirements for the particular application of Microsoft SQL Server 2000 Availability Does the solution have the necessary architec ture to ensure adequate failure protection such as built in redundancy of components data striping etc Recoverability In the event of a failure or corr
34. rver service Interlink performed this step to ensure that SQL caching did not influence test results Table 4 Resiliency Factors Tested Test Resiliency Factors Tested or introduced faults Description of Factors Tested 1 Remove one hard drive from the database drive set and This will introduce a degraded disk performance condition After a little time introduce the re introduce failed drive during SQL operation failed drive back into the array to let the system rebuild the array during the SQL operation 2 Remove one hard drive from the transaction log drive set This will introduce a degraded disk performance condition After a little time introduce the and introduce failed drive during SQL operation failed drive back into the array to let the system rebuild the array during the SQL operation 3 Unplug one network cable from the Snap Server 18000 This will reduce the network bandwidth available for database operations and is used to simu system while the database testing is in process late a failed network interface card NIC When complete repair the network cable by plugging the cable back in to the Snap Server 18000 system head unit This will restore the network subsystems to pre test condition 4 Unplug one network cable from the Snap Server 18000 This will reduce the network bandwidth available for disk rebuild operations and is used to system while the disk set is in the rebuild process simulate a failed
35. s that were file based iSCSI brings block level access and high performance to a lower cost storage medium This presents a compelling reason for IT to evaluate the potential for iSCSI as primary storage for database applica tions 1 1 Document Overview This document is intended to provide Adaptec with hard data from lab testing as to how the Adaptec iSCSI device compares against direct attached local SCSI disk storage in Microsoft SQL Server 2000 applications Interlink has provided this information within this document in the following format Solution Overview A general understanding of the bene fits and issues associated with using NAS devices as Microsoft SQL Server 2000 primary storage Testing Methodology How Interlink approached a gener alized performance characterization of Microsoft SQL Server 2000 Test Results A review of all test results with interpretation of findings Recommended Architecture Defines a high level architec ture based on testing results along with recommendations for extending this architecture in an Exchange production environment Conclusion A summary of findings and recommenda tions INTERLINK 1 2 Audience This document assumes the reader has a moderate under standing of networking and server environment technology terms and acronyms The intended audience is Adaptec cus tomers IT managerial and technical personnel 2 Solution Overview This section b
36. sing Windows Explorer rename the file to 5 file to DISBaseline DBSize blg DTSBaseline DBSize blg to ensure the proper perf mon file corresponds to the appropriate test Stop and Start Microsoft SQL Using Services stop and start the MSSQLServer Interlink performed this step to ensure that SQL caching did not 6 Server 2000 service influence test results Execute Test Script 1 Read to Using Microsoft SQL Server 2000 Query Analyzer This is a select statement on the targeted table e g 1GB 10GB 1 appropriate database run Select c unique c udist from examples or 100GB containing test data The result is that all rows in the table are selected and the columns returned are c unique and c udist Start Performance Counter Right mouse click on the saved log settings and Interlink started the perfmon after the test script was executed 8 choose Start to ensure no 0 values would occur in the test data 9 When test script ends stop per Right mouse click on saved log settings and choose Interlink stopped the perfmon log immediately after the test formance monitor stop script ended to ensure no O values would occur in the test data Rename Performance monitor log Using Windows Explorer rename the file to 10 file to ReadBaseline DBSize blg ReaadBaseline DBSize blg to ensure the proper perf mon file corresponds to the appropriate test 11 Stop and Start Microsoft SQL Using Services stop and start the SQLSe
37. so recommends that customers purchase an iSCSI TOE TCP IP Offload Engines card for database servers that will have iSCSI attached storage This will keep the server s CPU utilization from increasing when moving to this solution especially for database applications where the server s CPU uti lization may already be a point of concern Software based iSCSI initiators like Microsoft s freely downloadable iSCSI Software Initiator do work over standard NICs however they require the server s CPU to process the incoming and outgoing network stack tasks of the iSCSI protocol s packets A TOE card will take care of this processing and thus relieve the serv er s CPU of this task A TOE card will not improve network performance above its rated capacity however Realistically a TOE card and a standard 1000Mbps NIC will only send data as fast as the network will allow which is 1000Mbps or 100Mbps depending on the speed of the network devices used in the architecture Adaptec has certified some TOE cards and they are listed at http www snapappliance com page cfm name iscsi_initia tors nav technav Microsoft s iSCSI Software Initiator can be downloaded at http www microsoft com windowsserversystem storage tech nologies iscsi default mspx 13 SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION 5 3 3 Database Architecture Based on the performance characterization tests and the resiliency tests Interlink recommends p
38. that performed synthetic testing Interlink decided to use testing tools that performed actual tests on real Microsoft SQL Server 2000 databases These tools included DBGen To better understand read performance Interlink focused on disk throughput Interlink leveraged the Microsoft SQL Server 2000 Resource Kit DBGen tool to generate random data for a 1GB database 3 5 million row table a 10GB database 35 million row table and a 100GB database 350 million row table SQL Scripts With the data randomly created for the three databases the first test used a basic SOL select statement to test disk throughput as read operations Since the databases SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION did not have any indexes created on the columns all read operations were complete table scans and provided a good measure of disk throughput To test write performance under heavy conditions and understand drive performance and throughput to the Snap Server 18000 Interlink used DTS to copy in bulk a 3 5 million row 35 million row and 350 million row tables for each test configuration Perfmon Counters Interlink used the built in Perfor mance Monitor utility in Windows Server 2003 to moni tor log and average various system performance counters including CPU parameters physical disk performance net work performance and Microsoft SOL Server 2000 per formance Section 3 3 further describes
39. uption can data be quickly recovered Flexibility Does the solution offer flexibility in terms of adding capacity modifying storage allocation accommo dating multiple different RAID sets etc Manageability Can the device be easily managed and administered without complex knowledge requirements 2 2 Solution Description The Snap Server iSCSI solution is an attractive alternative to other storage scenarios for the following reasons Adaptability The use of network attached storage can help IT reduce isolated server centric disk storage that results in islands of information Administrators can quickly modify space allocation based on changing business requirements without having to perform complex and risky data backup and restoration Storage administrators can allocate a single device for multiple usage scenarios e g SQL Server 2000 ISCSI Snap SERVER 18000 CHARACTERIZATION file level access for file and print or block level access for database and messaging applications High Capacity vs Cost Adaptec offers terabytes of capaci ty at reasonable price points that create opportunities for storage optimization and centralization in organizations that were previously unable to afford such solutions Rapid Recovery Adaptec has bundled industry known software products with its offerings for rapid data recovery Its support for industry standards in file access NFS and CFS and emerging block leve
40. ver 2000 server s to the NAS over the public net work could result in network performance issues and unsatis INTERLINK factory Microsoft SQL Server 2000 performance Interlink recommends that customers treat iSCSI attached storage simi lar to Fibre Channel attached storage and create a dedicated storage backbone of Gigabit copper to isolate the traffic We make this recommendation for three reasons 1 Isolating this traffic will help to keep all the other traffic on the public network from interfering with Microsoft SQL Server 2000 performance 2 An organization s public network may not have implement ed 1000Mbps gigabit copper connections in its switch hard ware The Snap Server 18000 comes standard with two built in 10 100 1000Mbps gigabit copper connectors Utilizing Adaptec s fastest available connection speed will help ensure that undesirable Microsoft SQL Server 2000 performance is not introduced while implementing or migrating to the Snap Server 18000 for consolidated Microsoft SQL Server 2000 database storage 3 The NIC Network Interface Card in the Microsoft SOL Server 2000 may be busy enough already Introducing more network traffic on an already busy NIC may intro duce unsatisfactory client server performance The goal here is to ensure the highest throughput possible between the server and the iSCSI storage on the Snap Server 18000 5 3 2 Install iSCSI TOE Card Interlink al
Download Pdf Manuals
Related Search