Dell PowerEdge R720 SQL OLTP Virtualization Study

Dell PowerEdge R720 SQL OLTP Virtualization Study Measuring performance and power improvements of the new Intel Xeon processor E5-2600 product family Don Hoffman Solutions Performance Analysis

This document is for informational purposes only and may contain typographical errors and technical inaccuracies. The content is provided as is, without express or implied warranties of any kind. 2012 Dell Inc. All rights reserved. Dell and its affiliates cannot be responsible for errors or omissions in typography or photography. Dell, the Dell logo, EqualLogic, PowerConnect, and PowerEdge are trademarks of Dell Inc. Intel and Xeon are registered trademarks of Intel Corporation in the U.S. and other countries. Microsoft, Windows, SQL Server, and Windows Server are either trademarks or registered trademarks of Microsoft Corporation in the United States and/or other countries. VMware, ESXi, vcenter, and vsphere are either trademarks or registered trademarks of VMware Corporation in the United States and/or other countries. Other trademarks and trade names may be used in this document to refer to either the entities claiming the marks and names or their products. Dell disclaims proprietary interest in the marks and names of others. March 2012 Rev 1.0 ii

Contents Executive summary... 4 Introduction... 4 Test description... 5 Sample environment characterization... 6 Virtualized SQL system configurations... 7 Test methodology... 8 Overview... 8 Data collection overview... 9 Test results... 9 Virtual machine scaling to determine optimal load for each system... 9 Maximum Performance and System DBPM BIOS mode in the R720... 10 16 DIMM and 12 DIMM performance of the PowerEdge R720... 12 Direct performance and power comparison between the Dell PowerEdge R720 and PowerEdge R710 14 Idle power draw comparison... 14 Total operations per minute comparison... 14 Processing efficiency comparison... 15 Conclusion... 16 Appendix A: References... 17 Appendix B: Additional configuration details... 18 Tables Table 1. DVD Store database size description... 6 Table 2. System test configurations... 7 Table 3. Supporting infrastructure configurations... 8 Table 4. Performance and power draw of each BIOS mode... 11 Table 5. PowerEdge R720 16 DIMM and 12 DIMM comparison... 13 Table 6. Detailed system test configurations... 18 Table 7. Additional infrastructure and BIOS configurations... 19 Figures Figure 1. DVD Store test environment... 7 Figure 2. OPM scaling by number of VMs per SUT... 10 Figure 3. PowerEdge R720 Maximum Performance and System DBPM total OPM... 11 Figure 4. PowerEdge R720 Maximum Performance and System DBPM Performance per Watt... 12 Figure 5. PowerEdge R720 OPM per Watt 16 DIMM and 12 DIMM... 13 Figure 6. PowerEdge R720 and R710 idle power draw (lower is better)... 14 Figure 7. PowerEdge R720 and R710 total OPM... 15 Figure 8. PowerEdge R720 and R710 processing efficiency (OPM/watt)... 16 iii

Executive summary Dell s Solutions Performance Analysis team conducted a study to measure the performance and the performance per watt in a virtualized environment to compare the Intel Xeon processor 5600 series on the Dell PowerEdge R710 against the new Intel Xeon processor E5-2600 product family on the Dell PowerEdge R720. The configurations chosen were aimed at maximizing the performance of each system by equipping them with the top speed processors available, a 10Gb network connection to a backend SAN, and sufficient memory to run the workload at 100% processor utilization for both processor architectures tested. The study indicates that the Intel Xeon E5-2600 processors offer significant performance gains while doing so with higher processor efficiency (performance per watt).the Intel Xeon E5-2600 processors are also able to scale out further, supporting a higher number of virtual machines (VMs) at optimal performance. Key findings of the study include: The Dell PowerEdge R720 with Intel Xeon E5-2690 processors provided 53% higher virtualized online transaction processing (OLTP) performance than the PowerEdge R710 with Intel Xeon X5690 processors. The R720 configuration with Intel Xeon E5-2690 processors obtained 371,743 orders per minute (OPM) in a virtualized environment, as compared to the R710 configuration with Intel Xeon X5690 processors, which scored 242,622 OPM. The Dell PowerEdge R720 with Intel Xeon E5-2690 processors obtained 26% higher performance per watt than the PowerEdge R710 with Intel Xeon 5690 processors. The R720 configuration with Intel Xeon E5-2690 processors obtained a performance per watt score of 907 in a virtualized environment, as compared to the R710 configuration with Intel Xeon 5690 processors, which achieved a performance per watt score of 718. The Dell PowerEdge R720 with Intel Xeon E5-2690 processors used 26% less overall power at idle than the PowerEdge R710 with Intel Xeon X5690 processors. The R720 with Intel Xeon E5-2690 processors used an average of 136W at idle, as compared to the R710 configuration with Intel Xeon X5690 processors which used an average of 185W. Introduction In order to provide a better understanding for virtualization performance, this study describes specific application performance in a virtualized environment. The servers used in this study are the Dell PowerEdge R720 with Intel Xeon E5-2600 series processors, and the previous-generation Dell PowerEdge R710 with Intel Xeon X5600 series processors. Many other virtualization benchmarks exist, such as VMmark and SPECvirt, but this study is intended to answer a common question that cannot be answered by other benchmarks: What is the performance of a virtualized Microsoft SQL Server scenario on a mainstream dual-socket Dell server? 4

Considerations include: Comparing Microsoft SQL Server virtualized performance between the Dell PowerEdge R720 with Intel Xeon E5-2600 processors against the previous-generation Dell PowerEdge R710 equipped with Intel Xeon X5600 processors. The optimal number of VMs for each system will be determined by calculating the total OPM while scaling out the number of VMs assigned to each server. Comparing the power utilization of the Dell PowerEdge R720 with Intel Xeon E5-2600 processors against the Dell PowerEdge R710 with Intel Xeon X5600 processors at idle and 100% CPU utilization levels. Comparing the performance per watt of the Dell PowerEdge R720 with Intel Xeon E5-2600 processors against the Dell PowerEdge R710 with Intel Xeon X5600 processors. Comparing the power utilization, performance, and performance per watt of the Dell PowerEdge R720 with Intel Xeon E5-2600 processors when running in System DBPM (demandbased power management) BIOS profile against the same system running in Maximum Performance BIOS profile. Comparing the power utilization, performance, and performance per watt of the Dell PowerEdge R720 with Intel Xeon E5-2600 processors between 16 DIMM and 12 DIMM configurations. This study is not intended to be a comprehensive analysis of virtualization performance between Dell PowerEdge platforms; only the R720 and R710 were chosen to characterize the relative gains in performance and power efficiency achieved in the current generation of Dell servers. In addition, because ESXtop was actively logging performance counters during these runs, the numbers published in this study are not directly comparable to other published DVD Store results by either Dell or its competitors. Test description DVD Store Version 2 (DS2) is a complete online e-commerce test application with a back-end database component, a Web application layer, and driver programs. The goal in designing the database component as well as the mid-tier application was to utilize many advanced database features (for example, transactions, stored procedures, triggers, and referential integrity) while keeping the database easy to install and understand. The DS2 workload may be used to test databases or as a stress tool for any purpose. The distribution includes code for Microsoft SQL Server, Oracle, and MySQL databases. Included in the release are data generation programs; shell scripts to build data for 10 MB, 1 GB, and 100 GB 5

versions of the DVD Store; database build scripts and stored procedures; PHP Web pages; and a C# driver program. 1 Database workload sizes can be small, medium, or large, depending on environment setup. See Table 1 for details. Table 1. DVD Store database size description Database Orders Size Products Customers Small 10 MB 20,000 1,000/month 10,000 Medium 1 GB 2,000,000 100,000/month 100,000 Large 100 GB 200,000,000 10,000,000/month 1,000,000 Sample environment characterization To measure consolidated VM infrastructure host performance in a simulated e-commerce scenario, a full end-to-end solution was implemented. This environment consists of a physical server (host) running VMware ESX 4.1. This is referred to as the system under test (SUT) in the diagram. In addition to the SUT, the environment contains an environment controller system and a workload driver system. The environment controller is running Windows Server 2008 SP1. The intent of this system is to manage the VMware vcenter and vsphere client and act as a logging device for the power analyzer. The workload driver system is running Windows Server 2008 R2 Enterprise. This system is used to create simultaneous SQL user workload instances on the VMs that are being tested. In addition to these three physical systems, the environment contains a 10Gig Ethernet Dell PowerConnect 8024F switch, four Dell EqualLogic PS6010 SANs, and a Yokogawa WT210 power analyzer. The PowerConnect switch enables communications between the VMs and controller systems (workload driver and environment controller), and between the system under test (SUT) and the iscsibased EqualLogic storage backend. The power analyzer is connected to the environment controller system through a serial cable. The power analyzer takes one-second snapshots of all power readings and the controller logs this data during both the idle and active portions of the benchmark runs. 1 Test description information from Dell Enterprise Technology Center website: http://en.community.dell.com/techcenter/extras/w/wiki/dvd-store-readme.aspx 6

Figure 1. DVD Store test environment Virtualized SQL system configurations Configurations used in the test environment are shown in Table 2 and Table 3. Table 2. System test configurations Platform PowerEdge R720 PowerEdge R710 Operating system VMware ESXi 4.1 Update 2 Processor model 2 x Intel Xeon E5-2690 2 x Intel Xeon X5690 Processor frequency 2900 MHz 3467 MHz Memory details 12 x 8GB 2Rx4 LV RDIMMs 12 x 8GB 2Rx4 LV RDIMMs Memory frequency 1333 MT/s 1333 MT/s Internal storage 2 x 2.5 300GB 10K rpm SAS (RAID 1) BIOS power management setting Maximum Performance and System DBPM were tested 7

Platform PowerEdge R720 PowerEdge R710 Physical cores 16 12 Logical processors 32 24 # 2-vCPU VMs (optimal OPM) 16 6 Table 3. Environment controller system Supporting infrastructure configurations SQL workload driver system PowerEdge 1950 Windows Server 2008 Enterprise SP1 2 x Intel Xeon Processor E5410, 2.33 GHz Memory: 16 GB RAM Disk configuration: 2 x 73 GB SAS (RAID 1) Networking PowerConnect 8024F running at 10GbE 1 network power used per EqualLogic SAN (4 total) 1 network power used for SUT 1 port used as management port (via vsphere) Jumbo frames enabled on switch and SANs PowerEdge R905 Windows Server 2008 R2 Enterprise 4 x AMD Opteron 8435, 2.60 GHz Memory: 96 GB RAM Disk configuration: 2 x 146 GB SAS (RAID 1) Power analyzer Monitors SUT only Measures all power data during active portion of the benchmark Average watts during active and idle periods measured during official run Test methodology Overview The following methodology was used in the testing procedure: Determined the optimal number of virtual machines that each SUT can support. Each SUT was tested with 6, 12, 16, 18, and 20 VMs and the total OPM recorded. The number of VMs that provided the higher total OPM was the number chosen to test that particular server for the remainder of the study. Exercised all enabled VMs for a 10-minute run time. Recorded Performance metric, SUT Performance counters, and Power Measurements. Note: The performance metric recorded is OPM (orders per minute a measure of the number of orders processed during the benchmark runtime). 8

Data collection overview The following steps were used in collecting data: 1. Determined the number of VMs to enable during testing by scaling the number of VMs from 6 to 20 and recording the total OPM for each level. 2. Built a medium-size SQL database on each VM using DVD Store scripts. 3. Backed up each SQL database to a restore file, allowing the database to be restored from a pristine state, and then deleted all VM databases. 4. Restored all databases from the backup restore file and rebooted all VMs. 5. Performed a warm-up run to prepare databases for the official run, allowing each database to warm up individually and allowing a 10-minute idle period after warm up to allow VMs to return to a steady state. 6. Performed the official run starting all VMs simultaneously using a script. The power analyzer began measuring system power at the start of the official run and an ESX host was triggered to start ESXtop batch session during the official run. (Note: The VMware ESXtop tool gathered system-related performance counters for profiling CPU usage during official run.) 7. Checked the official run data for performance stability, collecting data if stable, and restarting the process if high variation was present. Data collected during the official run included: Average watts used during the official run (power metric) Average of the total ESXtop CPU counters during the official run (CPU percentage) VM throughput in OPM; the metric used in this study is an aggregate OPM measurement of all enabled VMs and is represented as performance in the test results Test results Virtual machine scaling to determine optimal load for each system In comparing the Dell PowerEdge R720 with the previous-generation Dell PowerEdge R710, it was necessary to determine the optimal loading of VMs on each system due to the architectural differences in the processors. The R720 features the new Intel Xeon E5-2600 series processors while the R710 comes equipped with the Intel Xeon 5600 series processors. The top speed E5-2690 processor has 8 physical processing cores with 20 MB L3 cache while the top speed X5690 processor has 6 physical processing cores with 12 MB L3 cache. The first test in this study was to determine how many VMs to assign each server for the remainder of the testing. To allow each system to perform at its highest level for the scaling study, each system was set to Maximum Performance mode in the BIOS and configured with 2 DIMMs per memory channel. The Intel Xeon X5690 processor has 3 memory channels per processor, where the Intel Xeon E5-2690 processor has 4 memory channels per processor. For a dual processor configuration, the R720 is optimally configured with 16 DIMMs and the R710 is optimally configured at 12 DIMMs. Later in this study, the 9

effects of setting the system performance mode to System DBPM and reducing the R720 DIMM count to be equal to the R710 is investigated. The results of this scaling study (illustrated in Figure 2) show that the R710 produces a maximum OPM (244,349) at 6 VMs and the R720 a maximum OPM (385,780) at 16 VMs. This optimal number of VMs will be used for the rest of this study for each system. Figure 2. OPM scaling by number of VMs per SUT This first comparison highlights a very important feature of the new Dell PowerEdge R720. With more and higher performing processing cores, the PowerEdge R720 is more scalable and can support a greater number of heavily loaded VMs without sacrificing individual VM performance. Maximum Performance and System DBPM BIOS mode in the R720 Maximum system performance is not the only consideration in this head-to-head comparison. System power draw is also an important factor in customer experience. Significant engineering resources go into optimizing the power and performance of each of the Dell PowerEdge BIOS profiles. This next study investigates the total performance (OPM) and average active power draw (Watts) of the Maximum Performance and System DBPM BIOS profiles. The processing efficiency of the system, measured in Performance per Watt (OPM/Watt), is a very important measurement that considers the power usage of the system, not just raw performance. The following comparison (shown in Figure 3) between Maximum Performance and System DBPM modes on the R720 was run with 16 VMs, the optimal number for this server. 10

Figure 3. PowerEdge R720 Maximum Performance and System DBPM total OPM Only 3.0% OPM difference between modes This first comparison shows that there is only a 3.0% performance difference between the two BIOS modes. This is very significant when considering the power savings that are possible when using the System DBPM mode. The next section of this study measures the average active power draw during the DVD Store runs for both BIOS modes and computes which mode provides the highest processing efficiency running this workload. Table 4. Performance and power draw of each BIOS mode PowerEdge R720 Maximum Performance System DBPM % difference OPM 385,780 374,339-3.0% Average power 455.7 W 425.6 W -6.6% OPM per watt 846.6 879.6 +3.9% 11

Figure 4. PowerEdge R720 Maximum Performance and System DBPM Performance per Watt System DBPM gives +3.9% OPM/Watt improvement While there is a 3% drop in overall OPM performance, the power draw drops by 6.6% when moving to System DBPM mode. This significantly lower power draw results in a +3.9% improvement in overall processing efficiency. System DBPM provides the best performance per watt out of the available BIOS modes. Because we want to provide the best possible performance per watt scores the System DBPM BIOS mode will be used for both servers for the remainder of the comparisons. 16 DIMM and 12 DIMM performance of the PowerEdge R720 In comparing the PowerEdge R720 against the PowerEdge R710, there is a disparity in the number of DIMMs in the system when optimally populated. Intel architectures give the best performance with two DIMMs per memory channel on each processor. Because of this, the PowerEdge R720 was originally tested with 16 DIMMs and the PowerEdge R710 with 12 DIMMs. Because this study also considers power consumption, it would be unfair to populate the R720 with additional DIMMs and raise its power draw artificially. This next study analyzes the overall performance, power draw, and performance per watt of the R720 when configured with 16 and 12 DIMMs. This study is done with the System DBPM BIOS profile, which was proven to give the best Performance per watt. 12

Table 5. PowerEdge R720 16 DIMM and 12 DIMM comparison PowerEdge R720 16 DIMMs 12 DIMMs % difference OPM 374,339 371,743-0.7% Average power 425.6 W 409.9 W -3.7% OPM per watt 879.6 906 +3.1% Figure 5. PowerEdge R720 OPM per Watt 16 DIMM and 12 DIMM 12 DIMM configuration provides +3.1% OPM/watt improvement There is a negligible 0.7% drop in overall OPM performance when configuring the R720 with 12 DIMMs instead of 16 DIMMs. The average active power draw drops by 3.7% and this increase in efficiency translates into a 3.1% improvement in the OPM/watt metric. This comparison proves that the R720 can handle unbalanced memory configurations very well and customers can be confident in scaling the installed memory to the exact capacity they need. This is the last data point needed in selecting the optimal PowerEdge R720 configuration. The rest of this study focuses on the Dell PowerEdge R720 and the PowerEdge R710 in equivalent configurations that will allow for direct comparison. 13

Direct performance and power comparison between the Dell PowerEdge R720 and PowerEdge R710 Idle power draw comparison The idle power usage of the two systems is the first comparison to consider. Both the PowerEdge R720 and PowerEdge R710 systems were set to System DBPM mode in the BIOS and both systems had 12 x 8GB 1333 MT/s LV RDIMM memory configurations. The power measurements were taken when the systems were at complete idle, having set inactive for 10 minutes between runs. The PowerEdge R720 server drew 136.2W at system idle and the R710 drew 185W at system idle. The lower power draw is better and the PowerEdge R720 consumed 26% less power at idle than the PowerEdge R710. Figure 6. PowerEdge R720 and R710 idle power draw (lower is better) The PowerEdge R720 uses 26% less power at idle than R710 Total operations per minute comparison The Dell PowerEdge R720 features the Intel Xeon E5-2600 processor family, which has a greater number of faster cores than the previous generation PowerEdge R710 with the Intel Xeon 5600 series processor. The additional performance capacity allows the R720 to support more VMs than the R710 and to operate them at a faster rate, which was shown in the initial VM scaling study in this white paper. This advantage is clearly shown when comparing the total OPM achieved by the R720 and R710 when running at the optimal number of VMs for each system s processor architecture. The R720 was able to 14

process 371,743 OPM compared to the R710 which was able to process 242,622 OPM. The Dell PowerEdge R720 was able to process 53% more OPM than the PowerEdge R710, a substantial improvement over just one generation of servers. Figure 7. PowerEdge R720 and R710 total OPM 53% greater OPM with the Dell PowerEdge R720 Processing efficiency comparison Considering total active power draw is the variable to consider in this server comparison. Not only do customers want a system that performs at a high level, but also one that does so while drawing the least amount of power possible. Calculating the total OPM/watt shows the measure of a system s processing efficiency. The formula for calculating OPM/watt is as follows: Aggregate performance (OPM) / Average power consumed = OPM per Watt (OPM/watt) The Dell PowerEdge R720 provided better Performance per Watt than the previous generation PowerEdge R710. The R720 scored 906.9 OPM/watt compared to the R710, which scored 717.8 OPM/watt. This gives the R720 an advantage of over 26% when compared to the R710. Not only does the R720 offer higher processing power than the R710, but it does so more efficiently as well. 15

Figure 8. PowerEdge R720 and R710 processing efficiency (OPM/watt) 26% greater performance per watt with the Dell PowerEdge R720 Conclusion The goal of this study was to characterize the performance and power impacts of the new Dell PowerEdge R720 server featuring Intel Xeon E5-2600 processors from the previous generation Dell PowerEdge R710 with Intel Xeon 5600 series processors using a virtualized database workload. Actual customer environments are unlikely to be running continuously at 100% processor utilization levels, but the relative scaling achieved with the transition to the PowerEdge R720 should still be indicative of the merits of the Intel Xeon processor E5-2600 product family and the overall design of the Dell PowerEdge R720. Based on the data contained in this study, the PowerEdge R720 with E5-2690 processors provides considerable performance and performance per watt advantages while reducing the overall system power at idle over the previous generation Intel Xeon processors in the PowerEdge R710. The PowerEdge R720 with Intel Xeon E5-2690 processors is the dual-socket platform of choice for maximum performance and scalability with additional processing cores and higher capacity memory support, as well as reduced overall power usage due to improved processing efficiency. For customers who desire a different form factor, this same architecture is also offered in a 1U server with the Dell PowerEdge R620 and in a tower configuration with the Dell PowerEdge T620. 16

Appendix A: References The DVD Store Virtualized SQL test methodology and overall structure of this white paper are due in large part to previous studies that were done before, as well as technical manuals and documents. Below are listed the key sources of information this study drew upon. Dell PowerEdge R910 SQL OLTP Virtualization Study: Measuring Performance and Power Improvements of New Intel Xeon E7 Processors and Low-Voltage Memory by Waseem Raja and John Beckett, attachments.wetpaintserv.us/jd7st0srn16bzholimozoq505396 Dell PowerEdge Intraline SQL OLTP Virtualization Study by Louis Barton and Greg Darnell. This is an internally published white paper not available publically online. Configuring VMware vsphere Software iscsi With Dell EqualLogic PS Series Storage, en.community.dell.com/dell-groups/dtcmedia/m/mediagallery/19861454.aspx How do I configure the optimal switch settings for IP based SAN? by Naveen Iyengar, en.community.dell.com/techcenter/enterprise-solutions/w/oracle_solutions/1422.aspx Dell EqualLogic Configuration Guide: A guide to building an iscsi based SAN solution with Dell EqualLogic PS Series Arrays, en.community.dell.com/techcenter/storage/w/wiki/equallogicconfiguration-guide.aspx 17

Appendix B: Additional configuration details Table 6. Detailed system test configurations Platform PowerEdge R720 PowerEdge R710 Processor model 2 x Intel Xeon E5-2690 2 x Intel Xeon X5690 Processor frequency 2900 MHz 3467 MHz Processor L3 cache 20 MB 12 MB Physical cores 16 12 Logical processors 32 24 Memory frequency 1333 MT/s 1333 MT/s Memory details 12 x 8 GB 2Rx4 LV RDIMMs 12 x 8 GB 2Rx4 LV RDIMMs Internal storage 2 x 2.5 300 GB 10K rpm SAS (RAID 1) HBA PERC H710p mini PERC H700 PCIe BIOS power management setting Maximum Performance and System DBPM were tested Physical cores 16 12 Logical processors 32 24 BIOS version 1.0.3 6.0.1 idrac version idrac7 Enterprise 1.0.0 (Build 69) idrac6 Enterprise 1.80.17 Total NIC ports used 1 x 10Gb PCIe, 1 x 1Gb LOM 1 x 10Gb PCIe, 1 x 1Gb LOM 10Gb Ethernet NIC 10Gb Intel XF PCI-E FC single port 10Gb Intel XF PCI-E FC single port 1Gb Ethernet onboard NIC Intel Gigabit 4P I350-t rndc Broadcom NetXtreme II (BCM5709C) Network firmware version 6.4.5 13.0.89 Power supply quantity and rating 1 x 870W 1 x 750W Line voltage used for testing 208V 208V 18

Table 7. Additional infrastructure and BIOS configurations Virtual environment details Platform PowerEdge R720 PowerEdge R710 VMware ESXi version 4.1.2 4.1.2 OS version installed per VM Windows 2008 Enterprise R2 x64 Windows 2008 Enterprise R2 x64 SQL version installed per VM Microsoft SQL 2008 R2 Microsoft SQL 2008 R2 Number of VMs (optimal configuration) 16 6 Logical processors per VM 2 4 EqualLogic configuration EqualLogic systems in use 4 4 EqualLogic system type EqualLogic PS6010 (RAID-10) EqualLogic PS6010 (RAID-10) Total storage per PS6010 3.66 TB 3.66 TB Size and # of HDDs/PS6010 16 x 600 GB 16 x 600 GB BIOS settings Turbo boost Enabled Enabled Processor prefetchers All enabled All enabled C-states Enabled Enabled C1E Enabled Enabled Logical processors Enabled Enabled Node interleaving Disabled Disabled Power management Maximum Performance and System DBPM both tested 19