IBM Storwize V7000 Gen2 Easy Tier with Microsoft Hyper-V, SQL, and Exchange Configuration and performance guidelines

Transcription

1 IBM Storwize V7000 Gen2 Easy Tier with Microsoft Hyper-V, SQL, and Exchange Configuration and performance guidelines David West IBM Systems, ISV Enablement July 2015 Copyright IBM Corporation, 2015

2 Table of contents Abstract... 1 Introduction... 1 IBM Storwize V7000 Gen IBM Easy Tier overview... 3 Version history... 3 Overview of the Storage Tier Advisor Tool... 4 Using the STAT program... 4 Using the STAT charting utility... 7 Testing... 9 Test environment... 9 Test tools...10 Configuration...10 SQL tests Exchange tests Test results...12 SQL Server...12 STAT charting utility output Exchange Server...16 STAT charting utility output Summary Resources Trademarks and special notices... 21

3 Abstract This paper covers the IBM Easy Tier automatic I/O hotspot migration feature with common Microsoft workloads on IBM Storwize V7000 Gen2. After testing with SQL and Exchange simulated workloads, the performance results and guidelines are considered. Easy Tier provides efficient I/O performance by dynamically keeping the most frequently accessed data on the fastest storage tier available. The paper is targeted at IT decision makers, systems implementers, and administrators with knowledge of SQL and Exchange storage concepts. Introduction Microsoft SQL and Exchange servers are critical business applications for many companies, potentially hosting a significant percentage of their data. This often results in huge data stores to house the information, however much of the older, or inactive data sits unused taking up expensive disk space. This is where tiered storage comes into play, and in the case of IBM Easy Tier, it is automated dynamic tiered storage. IBM Easy Tier 3, the latest version of Easy Tier (at the time of publishing this paper) in IBM Storwize software version 7.2 and later, can analyze and intelligently move data between three different tiers based on I/O hotspots or inactivity of the applications. For the more inactive or historical data, it can stay on lower cost disks, while the more frequently accessed data can move to either enterprise level serial-attached SCSI (SAS) or solid-state drives (SSD) storage. Easy Tier also offers an easy way to predict performance benefits before deployment, which is covered in the paper. Considering how important virtualization is in today s data centers, the project utilized Microsoft Hyper-V virtual machines (VMs) for the testing. The objective was to validate the workloads in a fully virtualized environment, although there was not a Hyper-V specific benefit that is tested with Easy Tier. As a result, this solution makes use of both virtual servers and storage for better efficiency in cloud and software defined environments. IBM Storwize V7000 Gen2 The IBM Storwize V7000 Gen2 storage system provides block storage that is enhanced with enterpriseclass features. Storwize V7000 Gen2 can now scale up to 20 expansion enclosures per control enclosure, and 504 drives, by using high-performance 12 Gbps SAS connections. Additionally, up to four control enclosures can be clustered, allowing the Storwize V7000 Gen2 system to scale both up and out. With the built-in storage virtualization, replication capabilities, and Microsoft storage integrations, the IBM Storwize V7000 system is a great fit for enterprise Microsoft workloads. 1

4 Figure 1: IBM Storwize V7000 overview IBM Storwize V7000 storage system enables customers to improve application flexibility, availability, responsiveness, and unmatched performance while reducing complexity and reducing storage space with the following features: Clustered systems: The clustering ability of Storwize V7000 system enables growth from the smallest configurations up to systems with 1056 drives for both performance and capacity. Data replication: Metro Mirror and Global Mirror perform synchronous and asynchronous data replication between Storwize V7000 systems at varying distances to protect data and keep services online in disaster situations. IBM Tivoli Storage FlashCopy Manager: The IBM Tivoli Storage FlashCopy Manager functionality enables the creation of instant volumes copies for data protection and flexibility. Advanced Storage tiering with IBM Easy Tier: Advanced technology for automatically migrating data between different storage tiers based on real-time usage and analysis. New Generation GUI: The graphical user interface (GUI) allows easy-to-use data management and point-and-click system management capabilities. IBM Storwize V7000 system is a modular storage system based on the IBM System Storage SAN Volume Controller (SVC) technology and uses the Redundant Array of Independent Disks (RAID) technology from the IBM System Storage DS8000 family to deliver a virtualized, easy to use, enterpriseready, mid-range storage solution. For additional information about the IBM Storwize V7000, refer to the following URL: ibm.com/systems/storage/disk/storwize_v7000/index.html 2

5 IBM Easy Tier overview The Storwize V7000 Easy Tier function provides a cost-effective, three-tiered SSD-enhanced solution. After Easy Tier function is enabled and performs an analysis of the data patterns for 24 hours, the most storage intensive data extents are identified then automatically migrated to higher-performing SSD or Enterprise level hard disk drive (HDD) arrays. The remaining cold data can take advantage of the highercapacity, lower-priced drives. Additionally, as the workload changes, cold data is demoted from higher to lower tiers. As a result, the applications experience the benefit of reduced latency and improved I/O and throughput without the cost of hosting the entire application on SSD storage. Each workload has its own unique I/O access patterns and performance characteristics over time. With only a small percentage of data relocation, a significant latency reduction can be achieved for the overall workload. Depending on the workload, larger SSD arrays can be required as well to provide the most benefit. By enabling Easy Tier analysis in the planning stages, the benefits of higher speed disks can be analyzed and predicted before purchasing. IBM offers the Storage Tier Advisor Tool (STAT), which reads the data activity and then generates reports and detailed recommendations. This allows IBM sales teams and IT organizations to plan ahead and set expectations for performance before implementation. There are numerous additional planning and performance tuning documents available through IBM websites. These documents delve into deeper planning, design, and technology overview topics. The documents are listed in the Resources section of this paper. Version history IBM Easy Tier was first introduced with the IBM System Storage DS8000 system, and the technology was later integrated into IBM Storwize V7000 in There have been multiple versions of Easy Tier for the DS8000 and the Storwize V7000 has implemented versions 1 and 3 to date, in Storwize family software version 7.4. The first generation of Easy Tier introduced automated tiering performance by intelligently moving data from enterprise-class drives to flash (SSD) drives. The technology that is allowed for limited flash deployments with minimal costs and large performance gains. This version introduced dynamic volume relocation and dynamic extent pool merging to accomplish the data migrations and balancing. The second generation was only implemented on DS8000. The third and current generation included in Storwize family software version 7.4 is referred to as Easy Tier 3, and now includes management across three tiers - SSD, enterprise-class, and near-line SAS drives. It supports single tier optimization, any combination of two drive types, or all three drive types. Also, new in version 3 is an improved Storage Tier Advisor tool, with enhanced reporting, graphing, and recommendation details. For single tier pools, Easy Tier does single tier optimization using automatic storage pool balancing to manage even performance across the managed disks (MDisks) in the pool. The storage pool balancing is part of the Easy Tier feature. 3

6 Overview of the Storage Tier Advisor Tool The Storage Tier Advisor Tool or STAT application interprets the workload data and provides a view of I/O performance and recommendations for improvement. It can be used before adding multiple tiers to a pool to predict and plan the optimal configuration before purchasing. It can also be used to monitor existing Easy Tier environments for optimization recommendations as workload demands change. Using the STAT program involves downloading a heat file from the Storwize V7000 support logs and loading the file into the program from the Microsoft Windows command line. After the tool reads the heat file, an HTML file is generated which can be viewed in a web browser. The STAT program can be downloaded here: There is also a very useful STAT charting utility, built with Microsoft Excel, which provides more detailed analysis and charts based on data files generated by the STAT program. The charting utility covers the following three key performance factors: Skew Chart shows workload activity by percentage of capacity Movement Chart shows Easy Tier data movement activity Workload Chart shows capacity utilization by extent pool The STAT charting utility can be downloaded here: Using the STAT program This section provides the steps to download the heat file from the Storwize V7000 and load it into the STAT program. 1. In the IBM Storwize GUI, click Settings in the left pane, and then click Support. 2. Click Show full log listing. 3. Click the node header at the upper-left corner of the window, and select the configuration node (node1 or node2). This is important because the heat file can be found only on the configuration node. 4. In the Filter field, type heat and press Enter for a list of heat files on the system, as shown in Figure 2. If you do not see any heat files, try checking the other node. 4

7 Figure 2: Location of heat files 5. Right-click the heat file and click Download to save the file to the STAT installation directory. 6. Use the following command in the Windows command line and STAT installation directory to load the heat file: C:\Program Files (x86)\ibm\stat>stat.exe -o c:\ex14 dpa_heat d data Note: The o option specifies the output path for the reports. 7. After running the command, go to the specified output path directory. 8. Open the index.html file to view the STAT web report. The following figures show the results from a sample Storage Tier Advisor Tool report. Figure 3 is a view of the heat distribution for the SQL database volume (VDisk ID 2) in monitoring mode before adding any SSDs to the system. Figure 4 shows the heat distribution after the SSD array is added to the pool and extent migration to SSD tier is in progress. 5

8 Figure 3: SQL database heat file during Easy Tier monitoring Figure 4: SQL database heat file after Easy Tier becomes active and data migrating to SSD array. The STAT tool also produces a skew chart of the monitored workload, which shows the weight of random and small-sized I/O activity. Figure 5 shows a sample view of the skew diagram that is generated for the tested SQL server TPC-C workload. The purpose of the skew chart is to show the distribution of small/random I/O over the tiers. In the chart in Figure 5, it is showing about 98% of the small/random I/O is on the SSD tier. 6

9 Figure 5: Skew chart for SQL TPC-C workload Figure 6 is a sample view of the SSD drive recommendations generated by the tool. In this example, the performance can be improved by adding more SSD drives, beyond the existing two 400 GB SSDs in the system. The performance improvements during testing and those predicted by the STAT program are directly related to the backend read and write latency reduction and not just host response time improvement. Figure 6: Recommended SSD configuration for SQL server workload Using the STAT charting utility The charting utility uses Microsoft Excel to read and work with the data. The following steps explain how to use the charting utility: 1. Open the Excel charting utility file named STAT Charting Utility_V1-5.xlsb. 2. Click on the Import Files button and select one or more of the three CSV files generated by the STAT program. The files are in the data_files directory of the STAT output folder as shown in figure 7. 7

10 Figure 7 The data_files directory where charting utility CSV files are located 3. Excel worksheets for each of the three chart type are created. Click on the tabs to view either detailed data or the charted data view as shown in figure 8. Figure 8 Sample view of charting utility reports 8

11 The charting utility reports and charts are reasonably self-explanatory. However for deeper analysis there is a reference worksheet included with the utility that explains the meaning of the parameters and values in more detail. Click on the first Excel tab to view the reference material. Testing The testing phase of the paper consisted of generating a simulated workload using well known stress testing tools for Microsoft workloads. Although no lab environment or simulated load can fully duplicate what occurs in a production environment, these tools are considered the closest to real-world loads as possible. The tests were configured large enough to generate sufficient workloads for demonstration purposes based on limited hardware and resources. Test environment To show an example of Easy Tier benefits with SQL server, an online transaction processing (OLTP) workload was used for the performance testing. OLTP workloads are characterized by high user concurrency rates and a pattern of high-rate small random I/O. This combination places a heavy demand on the storage system and generates the type of data Easy Tier detects for migration to higher tiers. I/O latency in OLTP environments is typically expected to be less than 10 milliseconds (ms) for read operations and 1 ms for write operations. Similarly, Exchange server has a heavy but larger random I/O pattern that expects database read latency well below 20 ms to ensure an adequate client experience. To meet these high I/O demands, a large number of spindles are normally required, and a high percentage of unused space. In these scenarios, Easy Tier can provide a performance boost by configuring a single SSD or Enterprise level storage array in the place of many slower spinning disks. The goal of the testing was to produce a reasonably large database that corresponds with the amount of SSD storage the test team had available. The configuration details are included below, however the test data set was not large enterprise sized due to resource limitations. From the test data, larger environments can use this data as a reference point to extrapolate benefits with larger configurations. The STAT tool is also available, as mentioned above, to predict and plan Easy Tier benefits in any size environment before deploying. Note: The lab tests only used two of the three tiers available, near-line SAS, and SSD drives. An ideal configuration is to have the third tier of storage, which consists of enterprise-level drives such as 10k or 15k SAS drives. The STAT tool analyzes all three tiers whether they are in the system or not, and reports the optimal number of Enterprises and or SSD drives recommended to improve the performance. The test objective was to show only a portion of the databases migrated to SSD, as a smaller subset of what a larger configuration would look like if more SSD drives were allocated with larger production databases. Migrating the entire database is not normally needed, since it is usually a smaller percentage of the data set that is heavily active and a candidate for migration. There can be a point of diminishing returns, where too many SSD drives do not help the overall performance or provide a good return on investment (ROI) due to the cost of the drives. Another aspect during SQL testing was to limit the amount of RAM available to SQL server to force more of a load on the storage system. This prevented the entire database from loading into server-side memory. 9

12 The test methods involved running the load tools and logging performance before and after Easy Tier monitoring windows and migration of extents to the SSD array. The tests were kept running for 2 to 3 days, providing a more realistic trend measurement over time, and also allowing time for hot data to migrate. After Easy Tier is enabled on a storage pool, it starts monitoring the data pattern and produces a heat file within 24 hours, followed by an updated heat file approximately every 24 hours. Test tools HammerDB is a load testing tool used for creating the SQL TPC-C database and generating sufficiently heavy I/O load to test Easy Tier functions. HammerDB is an open source database load testing and benchmarking tool for SQL Server, and many other popular databases. The tool includes complete built-in workloads based on industry standard benchmarks. The HammerDB workload is based on the TPC-C specification, and is designed to capture the essence of the TPC-C test without being an actual official TPC-C test result. As such it cannot be compared to official TPC-C submissions. However, it does an excellent job of creating reliable, consistent, and repeatable test results for comparing various database configurations. It requires a full installation of SQL Server, and the tool is normally run on a separate load server for best results, or even multiple load servers depending on the size of the tests. Microsoft Jetstress 2013 was used to test Exchange data because the tool has become the standard for evaluating and comparing storage I/O performance. Jetstress does not require an actual installation of Exchange server. It uses four Exchange DLL files, and is installed on each server to reliably simulate an Exchange I/O load. Configuration The largest SQL TPC-C database that you could configure with the preset HammerDB configurations was 5000 warehouses, which resulted in a 600 GB database during testing. By configuring a 400 GB SSD array, the team has provided the adequate space for hot data to migrate, without the entire database migrating. For Exchange, the test was configured with eight 488 GB databases, resulting in combined total database size of 3904 GB. The two 400GB SSD drives were configured as RAID 0 and an 800 GB array for this test, in order to provide more space. This is a significantly smaller ratio of about 1:5 SSD to database size compared to the SQL tests, however it still shows the excellent performance benefit after Easy Tier is enabled. The following are additional details of the server configurations used during the performance testing. 10

13 SQL tests The SQL tests used one large Hyper-V SQL virtual machine configured as follows: Microsoft Windows Server 2012 R2 MS SQL Server 2012 SP1 22 virtual processors 64 GB of RAM SQL server memory limited to 28 GB RAM Database and log volumes on pass-through FC connected disks Host bus adapter (HBA) queue depth at 64 IBM subsystem device driver device-specific module (SDDDSM) multi-path software Fifty-five 1 TB near-line SAS drives in one storage pool, RAID5 Two 400 GB SSD drives, RAID 1, added to same pool to enable Easy Tier TPC-C test with 5000 warehouses, resulting in a 600 GB database Note: The near-line SAS 7200 rpm drives are not the ideal choice for database loads, however the test team worked with what was available. An ideal test would include 10k or 15k SAS drives. The HammerDB load server consisted of one VM with the following configuration: Windows Server 2012 R2 four virtual processors 8 GB of RAM Exchange tests The Exchange test used one Hyper-V VM running Jetstress, configured as follows: Windows Server 2012 R2 12 virtual processors 64 GB RAM Database and logs on Fibre Channel (FC) connected pass-through disks HBA queue depth at 64 IBM SDDDSM multi-path software Eight 1 TB near-line SAS disks in one storage pool, RAID 5 Two 400 GB SSD drives, RAID 0 (to provide more SSD space for this test) Exchange 2013 SP1 DLL files Jetstress GB mailboxes, 2 database availability group (DAG) copies, eight 488 GB databases on four 1.5 TB volumes, 0.07 I/O operations per second (IOPS) profile 11

14 Test results This section covers the test results and shows the performance difference before and after migration of storage extents by Easy Tier. These configurations are not intended as production sizing guides for the given applications. The test was focused on creating a sufficient load on the storage system to observe the performance improvements per workload. SQL Server As illustrated in the following charts, the SQL server tests saw a 40 percent improvement in transactions per minute and 7 times faster disk read response. The disk write response time was nearly 100% improved. As a result, with Easy Tier enabled the disk latency for both read and write was at 1 millisecond. In this configuration, Easy Tier brought an overloaded SQL server workload closer to acceptable disk response times. The average disk queue length was three times lower with Easy Tier enabled, while the total I/O throughput per second increased nearly 20 percent. Figure 9: Improvement in TPC-C transactions per minute 12

15 Figure 10: Disk response time comparison with Easy Tier enabled Figure 11: Disk queue and throughput comparison Often a server s processor is under-utilized as it waits for I/O responses, which can result in wasted host resources. However, with Easy Tier s faster storage response times, server processor utilization improved nearly 100% as the storage drove more data at a faster rate. 13

16 Figure 12: SQL server processor utilization improvement STAT charting utility output The following charts present data from the three chart reports during peak activity of the SQL server tests. These are based on a heat file after most of the SQL hot data had moved to the SSD tier. Figure 13 Charting utility view of data movement as extents are moved to SSD tier 0 14

17 Figure 14 Charting utility view of workload skew by percentage of capacity Figure 15 Charting utility view of workload distribution between tier 0 and tier 1 15

18 Exchange Server Exchange 2013 tests on the Storwize Gen2 for this paper showed 47% less read latency. There was significant improvement in achieved IOPS as well, with 50% more. Exchange places a higher value on read response times; therefore, the primary improvement observed is on disk reads. Figure 16: Exchange disk latency comparison Figure 17: Achieved IOPS with Easy Tier enabled 16

19 STAT charting utility output The following charts present data from the three reports types during the Exchange tests. The workload is larger random I/O than SQL, which displays as warm and less active data. Figure 18 Charting utility view of data migration to SSD tier during Exchange testing Figure 19 Charting utility view of workload skew by percentage of capacity 17

20 Figure 20 Charting utility view of workload distribution between tier 0 and tier 1 For the SSD tier, Easy Tier considers small random I/O to be the most active and best candidate for SSD use. As a result it attempts to migrate this type of data first. For Exchange server, the data is larger and some of it more sequential, such as the background database maintenance that runs along-side the active and random load. Consequently, this mixed I/O pattern is viewed as less active; however it is still migrated to the higher tier, and provides performance improvement for the application. If the test environment had enterprise class drives installed as well, it is likely that Easy Tier would recommend a lot of this type of I/O for the enterprise tier. All IT environments are different, and as with any new solution, it is always recommended to test Easy Tier in each unique environment before deploying it in production. The STAT program can provide a wealth of valuable planning information. When testing Exchange for Easy Tier with Jetstress, ensure that the checksum process does not run at the end of the performance test. This can be avoided by checking the multihost check box in Jetstress, which stops the test after the performance run, and requires manual intervention to proceed with the checksum portion of the test. The checksum workload in Jetstress can run for many hours, and is a very different workload that will skew the Easy Tier workload monitoring. When running test workloads to determine Easy Tier recommendations, it is important to keep the test running at least 24 hours, or longer, to keep a consistent workload running and have accurate Easy Tier heat file results. If the tests stop for an extended period, there can be demotion of assumed cold data to lower tiers or result in unexpected workload evaluation. 18

21 Summary This paper covered Microsoft workloads with Easy Tier technology for efficient storage performance gains. By providing a relatively small SSD array, significant performance improvement can be achieved by moving only the most active data to the SSD tier. Easy Tier takes advantage of high performance SSD storage without requiring the entire workload on the higher priced platform. For further reading, additional in-depth information on Easy Tier is available in the Resources section. 19

22 Resources These websites provide useful references to supplement the information contained in this paper: IBM Systems on PartnerWorld ibm.com/partnerworld/systems IBM Redbooks ibm.com/redbooks IBM SVC and Storwize V7000 best practice guide Implementing the IBM Storwize V7000 Gen2 Combining Easy Tier with Real-time Compression IBM Storwize V7000 information center ibm.com/support/knowledgecenter/st3fr7/welcome IBM Publications Center 20

23 Trademarks and special notices Copyright IBM Corporation References in this document to IBM products or services do not imply that IBM intends to make them available in every country. IBM, the IBM logo, and ibm.com are trademarks or registered trademarks of International Business Machines Corporation in the United States, other countries, or both. If these and other IBM trademarked terms are marked on their first occurrence in this information with a trademark symbol ( or ), these symbols indicate U.S. registered or common law trademarks owned by IBM at the time this information was published. Such trademarks may also be registered or common law trademarks in other countries. A current list of IBM trademarks is available on the Web at "Copyright and trademark information" at Microsoft, Windows, Windows NT, and the Windows logo are trademarks of Microsoft Corporation in the United States, other countries, or both. Other company, product, or service names may be trademarks or service marks of others. Information is provided "AS IS" without warranty of any kind. All customer examples described are presented as illustrations of how those customers have used IBM products and the results they may have achieved. Actual environmental costs and performance characteristics may vary by customer. Information concerning non-ibm products was obtained from a supplier of these products, published announcement material, or other publicly available sources and does not constitute an endorsement of such products by IBM. Sources for non-ibm list prices and performance numbers are taken from publicly available information, including vendor announcements and vendor worldwide homepages. IBM has not tested these products and cannot confirm the accuracy of performance, capability, or any other claims related to non-ibm products. Questions on the capability of non-ibm products should be addressed to the supplier of those products. All statements regarding IBM future direction and intent are subject to change or withdrawal without notice, and represent goals and objectives only. Contact your local IBM office or IBM authorized reseller for the full text of the specific Statement of Direction. Some information addresses anticipated future capabilities. Such information is not intended as a definitive statement of a commitment to specific levels of performance, function or delivery schedules with respect to any future products. Such commitments are only made in IBM product announcements. The information is presented here to communicate IBM's current investment and development activities as a good faith effort to help with our customers' future planning. Performance is based on measurements and projections using standard IBM benchmarks in a controlled environment. The actual throughput or performance that any user will experience will vary depending upon considerations such as the amount of multiprogramming in the user's job stream, the I/O configuration, the storage configuration, and the workload processed. Therefore, no assurance can be given that an individual user will achieve throughput or performance improvements equivalent to the ratios stated here. 21

24 Photographs shown are of engineering prototypes. Changes may be incorporated in production models. Any references in this information to non-ibm websites are provided for convenience only and do not in any manner serve as an endorsement of those websites. The materials at those websites are not part of the materials for this IBM product and use of those websites is at your own risk. 22