White PapersWith rising energy costs continuing to demand larger portions of organisational budgets, Dell Inc conducted a range of tests to examine the true power efficiency of the top three global blades server providers and compare and contrast the results. Each blade configuration included an enclosure configured as similarly as possible and was fully populated with each company’s best-selling blade servers using identical processor, RAM and hard drive selections. The tests revealed that the performance and power efficiency of Dell M-series blades can result in significant operational cost savings.
Power Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures A Dell Technical White Paper ohn Beckett, Robert Bradfield, and the Dell Server Performance Analysis Team Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 2 THIS WHITE PAPER 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. 2010 Dell Inc. All rights reserved. Reproduction of this material in any manner whatsoever without the express written permission of Dell Inc. is strictly forbidden. For more information, contact Dell. Dell, the DELL logo, and the DELL badge, and PowerEdge are trademarks of Dell Inc. Microsoft, Windows, and Windows Server are either trademarks or registered trademarks of Microsoft Corporation in the United States and/or other countries. SPEC and the benchmark name SPECpower_ssj are trademarks of the Standard Performance Evaluation Corporation. Benchmark results stated in this white paper reflect results run as of July 6, 2010. The comparison presented in this paper is based on the enterprise-class blade servers configured as similarly as possible currently shipping by Dell, HP and IBM. For the latest SPECpower_ssj2008 benchmark results, visit http://www.spec.org/power_ssj2008/results/power_ssj2008.html. 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 Inc. disclaims any proprietary interest in trademarks and trade names other than its own. August 2010 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 3 Contents Executive Summary ....................................................................................................... 5 Introduction ............................................................................................................. 5 Key Findings ............................................................................................................. 5 Power .................................................................................................................. 5 Performance .......................................................................................................... 6 Testing Details ............................................................................................................. 6 Methodology ............................................................................................................. 6 Results .................................................................................................................... 6 Power Draw on a Per Blade Basis .................................................................................. 12 Workload ............................................................................................................... 14 Overall ssj_ops/watt ................................................................................................. 14 Appendix A Test Methodology ........................................................................................ 18 SPECpower_ssj2008 Standard ....................................................................................... 18 Configuration .......................................................................................................... 19 Chassis Configuration ................................................................................................ 19 BIOS Settings .......................................................................................................... 19 OS Tuning .............................................................................................................. 20 SPECpower_ssj2008 Configuration ................................................................................. 20 Power Meter Configuration ......................................................................................... 21 Appendix B Enclosure Configuration Information ................................................................. 24 Appendix C Blade System Configuration Information ........................................................... 25 Appendix D Blade System Firmware and Drivers ................................................................ 26 Appendix E SPECpower_ssj2008 Results ............................................................................ 27 Dell M1000e Blade Enclosure/16 x PowerEdge M610 ........................................................... 27 HP BladeSystem c7000/16 x ProLiant BL460c G6 ............................................................... 34 IBM BladeCenter 8852 H-Series/14 x BladeCenter HS22 7870 ................................................ 41 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 4 Tables Table 1. SPECpower_ssj2008 Results for the 16-Blade Dell Solution ....................................... 15 Table 2. SPECpower_ssj2008 Results for the 16-Blade HP Solution ........................................ 16 Table 3. SPECpower_ssj2008 Results for the 14-Blade IBM Solution ....................................... 17 Table 4. Configuration for Testing ............................................................................... 19 Table 5. Enclosure Configuration Information ................................................................. 24 Table 6. Blade System Configuration Information ............................................................. 25 Table 7. Detailed Configuration Information for Blade Server Solutions .................................. 26 Figures Figure 1. Blade Solution Comparison Chart ....................................................................... 7 Figure 2. Comparison of the Blade Solutions at the Active Idle Power Measurement ..................... 8 Figure 3. Power Usage Comparison of Blade Solutions at 100% Utilization .................................. 9 Figure 4. Average Power Utilization Across Workload Intervals ............................................. 10 Figure 5. Performance Comparison of Watt Ratios Across Workload Intervals ............................ 11 Figure 6. Performance Comparison at 100% Utilization ....................................................... 12 Figure 7. Comparison of the Power Utilization on a per Blade Basis ....................................... 13 Figure 8. Comparison of the Overall ssj_ops/watt Score ..................................................... 14 Figure 9. Power Measurement Diagram for the Dell Blade Solution ........................................ 21 Figure 10. Power Measurement Diagram for the HP Blade Solution ....................................... 22 Figure 11. Power Measurement Diagram for the IBM Blade Solution ...................................... 23 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 5 Executive Summary Introduction With rising energy costs continuing to demand larger portions of organizational budgets, Dell Inc. (Dell) commissioned its Server Performance Analysis team to use the industry standard SPECpower_ssj2008 benchmark to compare the power draw and performance per watt of blade solutions from Dell, HP, and IBM. The purpose of the testing was to examine on a level playing field the true power efficiency of the top three global blades server providers and compare and contrast the results. Each blade configuration included an enclosure configured as similarly as possible and was fully populated with each company s best-selling blade servers1 using identical processor, RAM and hard drive selections. The results were clear. In like-for-like comparisons, a Dell M1000e enclosure fully populated with 16 M610 blade servers consistently demonstrated significant advantages over both the IBM BladeCenter H enclosure fully populated with 14 HS22 blade servers and the Hewlett-Packard C7000 enclosure fully populated with 16 BL460C G6 blade servers in both power draw and performance-per-watt, regardless of utilization levels. While the actual monetary impact of these savings may vary for a variety of reasons, such as the regional disparity in cost of electricity and overall data center power and cooling design efficiency, there is no doubt that the performance and power efficiency of Dell M-series blades can result in significant operational cost savings. Key Findings Key findings from the study for power and performance are summarized below. Power The Dell M1000e enclosure fully populated with 16 M610 blade servers used less power across all load levels than either the HP C7000 enclosure fully populated with 16 BL460C G6 blade servers or the IBM BladeCenter H enclosure fully populated with 14 HS22 blade servers (Figure 4). In the CPU utilization midrange (40 60%), the HP C7000 enclosure fully populated with 16 BL460C G6 blade servers used 13 17% more power per server than Dell M1000e enclosure fully populated with 16 M610 blade servers and the IBM BladeCenter H enclosure fully populated with 14 HS22 blade servers used 19-20% more power per server (Figure 4). The IBM BladeCenter H enclosure fully populated with 14 HS22 blade servers used 63.6% more power at idle than the Dell M1000e enclosure fully populated with 16 M610 blade servers, despite having two fewer blade servers (Figure 2). A HP C7000 enclosure fully populated with 16 BL460C G6 blade servers used 24% more power than the Dell M1000e enclosure fully populated with 16 M610 blade servers at idle and 13% more at 100% CPU utilization (Figure 2 and Figure 3). In head-to-head testing with fully populated chassis, IBM HS22 blades used 87% more power per blade at idle than Dell PowerEdge M610 blade servers (Figure 7). In head-to-head testing with fully populated chassis, IBM HS22 blades used an average of 16.1% more power per blade at 100% CPU utilization than Dell PowerEdge M610 blade servers (Figure 7). 1 IDC Worldwide Quarterly Server Tracker Q1 CY2010 results Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 6 Performance The Dell M1000e enclosure fully populated with 16 M610 blade servers achieved a higher performance to power ratio across all load levels than the HP C7000 enclosure fully populated with 16 BL460C G6 blade servers or the IBM BladeCenter H enclosure fully populated with 14 HS22 blade servers (Figure 5). Despite drawing less power, the Dell M1000e enclosure fully populated with 16 M610 blade servers provided 13.9% greater performance than the IBM BladeCenter H enclosure fully populated with 14 HS22 blade servers at 100% utilization (Figure 6). The Dell M1000e enclosure fully populated with 16 M610 blade servers achieved up to 15% higher performance/watt than the HP C7000 enclosure fully populated with 16 BL460C G6 blade servers and up to 22% higher performance/watt than the IBM BladeCenter H enclosure fully populated with 14 HS22 blade servers (Figure 8). Test methodology and detailed results are documented in this paper. Testing Details Methodology SPECpower_ssj2008 is an industry standard benchmark created by the Standard Performance Evaluation Corporation (SPEC) to measure a server s power and performance across multiple utilization levels. Appendix A details the test methodology used by Dell, Appendices B D detail configuration for the tests, and Appendix E provides detailed report data that supports the Results in this paper. Results The Dell blade solution delivered the best SPECpower_ssj2008 result (see Figure 1), coming in at 2,530 overall ssj_ops/watt compared to the HP blade solution, which came in with 2,197 overall ssj_ops/watt (Dell 15% higher). The IBM blade solution came in last with 2,068 overall ssj_ops/watt (Dell 22% higher). This result is even more notable due to the fact that IBM is only capable of providing 14 blade servers per chassis, and thus is providing a solution that, despite drawing significantly more power, is not capable of the same raw performance per chassis that HP or Dell can provide using identical processor and memory architectures. Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 7 Figure 1. Blade Solution Comparison Chart 1.00 1.00 1.00 1.00 1.13 1.24 0.87 1.00 1.02 1.64 0.82 0.88 0.000.200.400.600.801.001.201.401.601.80100% avg power (loweris better)Active Idle Power(lower is better)SPECpower_ssj2008Score (higher is better)100% ssj_ops (higher isbetter)Normalized Blade Solution Comparison Power Utilization, Performance/Watt and Performance Dell blade solutionHP blade solutionIBM blade solutionUntitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 8 SPECpower_ssj2008 includes a measurement of power while the blades are idle at the final phase of the benchmark. As Figure 2 shows, the full enclosure of 16 Dell PowerEdge M610 blade servers used 24.1 percent less overall power while idle than the HP blade solution. In addition, the 16 Dell PowerEdge M610 blade servers used 63.6 percent less overall power while idle than the 14 IBM HS22 blade servers. The 16-blade Dell PowerEdge M610 solution used 1,288 watts at idle, while the 16-blade HP ProLiant BL460c solution used 1,598 watts at idle. The 14-blade IBM HS22 solution used 2,107 watts at idle. Figure 2. Comparison of the Blade Solutions at the Active Idle Power Measurement 1,288 1,598 2,107 05001,0001,5002,0002,500Dell 16-Blade SolutionHP 16-Blade SolutionIBM 14-Blade SolutionSPECpower_ssj2008 average watts @ active idle (lower watts are better) Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 9 SPECpower_ssj2008 includes a measurement of power while the blades are at 100% utilization. As Figure 3 shows, the full enclosure of 16 Dell PowerEdge M610 blade servers used 11.5 percent less overall power at 100% utilization than the HP blade solution and 1.6% percent less than the IBM BladeCenter H enclosure with 14 HS22 blade servers. The 16-blade Dell PowerEdge M610 solution used 4,372 watts at 100% utilization, while the 16-blade HP ProLiant BL460c solution used 4,940 and the 14-blade IBM HS22 solution used 4,444 watts (despite the fact that the IBM blade solution has 2 less blades than the Dell & HP blade solutions). Figure 3. Power Usage Comparison of Blade Solutions at 100% Utilization 4,372 4,940 4,444 4,0004,2004,4004,6004,8005,000Dell 16-Blade SolutionHP 16-Blade SolutionIBM 14-Blade SolutionSPECpower_ssj2008 average watts @ 100% workload (lower watts are better) Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 10 Figure 4 shows the average interval power in watts while running the workload at 100% target load down to 10 percent in stepped 10 percent utilization increments as reported by the SPECpower_ssj2008 benchmark. A full enclosure of 16 Dell PowerEdge M610 blade servers used less power than both the 16 HP ProLiant BL460c G6 blade servers and the 14 IBM HS22 blade servers across all of the SPECpower_ssj2008 target load levels. The power deltas were greater between the Dell blade solution and the HP blade solution, where at higher utilization levels (60% 100%) we saw a difference of between 13 and 18.9% in favor of the Dell blade solution. At the same utilization levels, the delta between the Dell blade solution and the IBM blade solution was between 1.6 and 5.4%. At lower utilization ranges of the SPECpower_ssj2008 benchmark (50% 10%) we saw a delta between the Dell blade solution and the HP blade solution of between 11.8 and 14.2%, and a delta of between 4.2 and 11.6% between the Dell blade solution and the IBM blade solution. Figure 4. Average Power Utilization Across Workload Intervals 01,0002,0003,0004,0005,0006,000100%90%80%70%60%50%40%30%20%10%ActiveIdleWorkload Percentile SPECpower_ssj2008 Average Power at Each Workload Percentile Dell Blade SolutionHP Blade SolutionIBM Blade SolutionUntitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 11 Figure 5 shows the performance-to-watt ratio results for the target load percentages. Results are the measured throughput (performance) divided by the average power consumption for each of the workload intervals. The 16 HP ProLiant BL460c G6 blade servers achieved almost identical 100% throughput to the16 Dell PowerEdge M610 blade servers , but used more power at this and every other workload interval. As a result, the Dell blade solution achieved a higher performance-to-watt ratio than the HP blade solution across all load levels. The 16 Dell PowerEdge M610 blade servers achieved 14% higher throughput than the 14 IBM BladeCenter HS22 Blade servers, and used less power across all workload intervals. As a result, the Dell blade solution also achieved a higher performance-to-watt ratio than the IBM blade solution across all load levels. Figure 5. Performance Comparison of Watt Ratios Across Workload Intervals 05001,0001,5002,0002,5003,0003,5004,000100%90%80%70%60%50%40%30%20%10%Target Load Percentages SPECpower_ssj2008 Performance To Watt Ratios Higher is better Dell Blade SolutionHP Blade SolutionIBM Blade SolutionUntitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 12 SPECpower_ssj2008 includes a measurement of performance at 100% utilization. As Figure 6 shows, the full enclosure of 16 Hewlett-Packard BL460c blades was almost equivalent in performance to the Dell PowerEdge M-Series blade solution. However, the 16 Dell PowerEdge M610 blade servers achieved 13.9% higher performance at 100% utilization than the 14 IBM HS22 blade servers. The 16-blade Dell PowerEdge M610 solution achieved 14,785,342 ssj_ops at 100% utilization, while the 16-blade HP ProLiant BL460c solution achieved 14,774,218 and the 14-blade IBM HS22 solution was able to reach 12,979,356 ssj_ops. Figure 6. Performance Comparison at 100% Utilization Power Draw on a Per Blade Basis Due to the inequities in the number of blades per chassis between the 14 Blade IBM solution and the 16 Blade Dell and HP solutions, we also looked at the power draw on a per blade basis by dividing the average power for each SPECpower load level by the number of blades in each chassis. We saw that the Dell blade solution was more power efficient on a per blade basis at every load level than either the HP blade solution or the IBM blade solution. The efficiency differences are especially striking when comparing the Dell blade solution against the IBM blade solution at the active idle workload level, where the IBM solution draws 87% more power at idle on a per blade basis as the Dell blade solution. The Dell blade solution calculated on a per blade basis drew 81W versus 151W for the IBM blade solution. The HP blade solution used 24.1% more power on a per blade basis than the Dell blade 14,785,342 14,774,218 12,979,356 12,000,00012,500,00013,000,00013,500,00014,000,00014,500,00015,000,000Dell Blade SolutionHP Blade SolutionIBM Blade SolutionSPECpower_ssj2008 ssj_ops @ 100% Workload Higher ssj_ops are better Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 13 solution, where the per blade power utilization was calculated to be 100W for the HP solution, and 81W for the Dell solution. In addition, when comparing the Dell blade solution against the IBM blade solution at the 100% load level on a per blade basis, the IBM blade solution drew 16.2% more power than the Dell blade solution. The Dell solution on a per blade basis was calculated to draw 273W versus 317W for the IBM solution. The HP blade solution drew up to 13% more power at the 100% load level than the Dell blade solution on a per blade basis, where the per blade power utilization was calculated to be 273W for the Dell blade solution versus 309W for the HP blade solution. See Figure 7. Figure 7. Comparison of the Power Utilization on a per Blade Basis 050100150200250300350100%90%80%70%60%50%40%30%20%10%ActiveIdlePower Draw per Blade Across All Load Levels Lower watts are better Dell Per BladeHP Per BladeIBM Per BladeUntitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 14 Workload During a SPECpower_ssj2008 run, the system under test runs three calibration phases to determine the 100% workload target ssj_ops. Once this value is determined, then the benchmark runs the 100% workload interval, collecting performance and power data, and then scales the workload back in 10% increments until reaching the active idle measurement interval. The active idle interval runs no workload on the server(s) under test, but the power is collected for 5 minutes and averaged. The final SPEcpower_ssj2008 score is the server s performance (in ssj_ops) summed across all workload intervals, divided by the average power summed across all workload intervals. A higher SPECpower_ssj2008 ssj_ops/watt score is better. Overall ssj_ops/watt Figure 8 shows the SPECpower_ssj2008 results for the Dell blade solution, the HP blade solution, and the IBM blade solution in overall ssj_ops/watt. Figure 8. Comparison of the Overall ssj_ops/watt Score 2,530 2,197 2,068 05001,0001,5002,0002,5003,000Dell Blade SolutionHP Blade SolutionIBM Blade SolutionSPECpower_ssj2008 Overall Score Comparison Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 15 Table 1 shows the SPECpower_ssj2008 results for the Dell PowerEdge solution for each target load. Table 1. SPECpower_ssj2008 Results for the 16-Blade Dell Solution Dell PowerEdge M610 (16 total blades) Performance Power Performance to Power Ratio Target Load Actual Load ssj_ops Average Active Power (W) 100% 99.70% 14,785,342 4,372 3,382 90% 90.00% 13,344,934 3,987 3,347 80% 79.90% 11,853,404 3,646 3,251 70% 70.00% 10,380,753 3,359 3,090 60% 60.00% 8,903,133 3,104 2,868 50% 50.00% 7,412,879 2,900 2,556 40% 40.00% 5,928,182 2,692 2,203 30% 30.00% 4,444,109 2,496 1,781 20% 20.00% 2,971,405 2,301 1,292 10% 10.00% 1,481,894 2,067 717 Active Idle 0 1,288 0 ssj_ops / power = 2,530 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 16 Table 2 shows the SPECpower_ssj2008 results for the HP blade solution for each target load. (Lower Watts are better. Higher ssj_ops are better.) Table 2. SPECpower_ssj2008 Results for the 16-Blade HP Solution HP Solution 16 total blades Performance Power Performance to Power Ratio Target Load Actual Load ssj_ops Average Active Power (W) 100% 99.60% 14,774,218 4,940 2,991 90% 90.00% 13,338,325 4,608 2,895 80% 80.00% 11,858,862 4,306 2,754 70% 70.00% 10,377,087 3,977 2,609 60% 60.00% 8,895,651 3,630 2,451 50% 50.00% 7,420,425 3,313 2,240 40% 39.90% 5,922,357 3,038 1,949 30% 30.00% 4,452,479 2,801 1,589 20% 20.00% 2,971,369 2,574 1,154 10% 10.00% 1,479,418 2,310 641 Active Idle 0 1,598 0 ssj_ops / power = 2,197 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 17 Table 3 shows the SPECpower_ssj2008 results for the IBM blade solution for each target load. Table 3. SPECpower_ssj2008 Results for the 14-Blade IBM Solution IBM HS22 (14 total blades) Performance Power Performance to Power Ratio Target Load Actual Load ssj_ops Average Active Power (W) 100% 99.60% 12,979,356 4,444 2,921 90% 89.90% 11,716,988 4,141 2,829 80% 80.10% 10,429,739 3,865 2,698 70% 70.10% 9,128,012 3,566 2,559 60% 60.00% 7,820,687 3,273 2,390 50% 50.00% 6,516,261 3,021 2,157 40% 39.90% 5,202,062 2,813 1,850 30% 30.00% 3,905,546 2,634 1,483 20% 20.00% 2,604,026 2,462 1,058 10% 10.00% 1,302,857 2,306 565 Active Idle 0 2,107 0 ssj_ops / power = 2,068 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 18 Appendix A Test Methodology SPECpower_ssj2008 Standard SPECpower_ssj2008 is an industry standard benchmark created by the Standard Performance Evaluation Corporation (SPEC) to measure a server s power and performance across multiple utilization levels. SPECpower_ssj2008 consists of a Server Side Java (SSJ) workload along with data collection and control services. SPECpower_ssj2008 results portray the server s performance in ssj_ops (server side Java operations per second) divided by the power used in watts (ssj_ops/watt). SPEC created SPEcpower_ssj2008 for those who want to accurately measure the power consumption of their server in relation to the performance that the server is capable of achieving with ssj2008 workload. SPECpower_ssj2008 consists of three main software components: Server Side Java (SSJ) Workload Java database that stresses the processors, caches and memory of the system, as well as software elements such as OS elements and the Java implementation chosen to run the benchmark. Power and Temperature Daemon (PTDaemon) Program that controls and reports the power analyzer and temperature sensor data. Control and Collect System (CCS) Java program that coordinates the collection of all the data. For more information on how SPECpower_ssj008 works, see http://www.spec.org/power_ssj2008/. All results discussed in this whitepaper are from compliant runs in SPEC terminology, which means that although they have not been submitted to SPEC for review, Dell is allowed to disclose them for the purpose of this study. All configuration details required to reproduce these results are listed in Appendices A, B, and C, and all result files from the runs compared are included in Appendix D. Each blade solution was configured by installing a fresh copy of Microsoft Windows Server 2008 Enterprise R2 on each blade with the operating system installed on a two-hard drive RAID 1 (or RAID 1 + 0 in the case of the HP blade solution) choosing the full installation option for each. The latest driver and firmware update packages available to all three blade solutions were installed at the beginning of this study. Refer to Appendix B for details. The Dell Server Performance Analysis Team ran SPECpower_ssj2008 three times per configuration across all three blade solutions and chose the highest overall ssj_ops/watt score for each configuration to compare for this study. Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 19 Configuration Table 4 details the configuration used for testing. Table 4. Configuration for Testing Configuration Dell Blade Solution HP Blade Solution IBM Blade Solution Blade Chassis PowerEdge M1000e BladeSystem c7000 BladeCenter H-series Blade Type M610 BL460c HS22 # Blades 16 16 14 (maximum possible) CPU Type per blade 2 x X5670 2.93GHz 2 x X5670 2.93GHz 2 x X5670 2.93GHz Memory per blade 6 x 4GB 1333MHz DDR3 6 x 4GB 1333MHz DDR3 6 x 4GB 1333MHz DDR3 HDD per blade 2 x 73GB 15K SAS 2 x 73GB 15K SAS 2 x 73GB 15K SAS Operating System Microsoft Windows Server 2008 Enterprise R2 Microsoft Windows Server 2008 Enterprise R2 Microsoft Windows Server 2008 Enterprise R2 Java Version Used IBM J9 052192009 IBM J9 052192009 IBM J9 052192009 Chassis Configuration The team configured the three blade solution chassis for AC Redundancy where applicable, and left any Dynamic Power Supply Engagement options to the default settings. For the Dell PowerEdge M1000e chassis, the Power Redundancy mode defaulted to AC Redundancy, and the Dynamic Power Supply Engagement Mode defaulted to Disabled. For the HP c7000 Blade Enclosure, we selected AC Redundancy, and left the Dynamic Power option at the default of Enabled2. For the IBM BladeCenter H-Chassis configuration, we changed the Power Management mode from Basic to Redundant Power Management, which best matched our Redundancy choices for the HP and Dell blade solutions. BIOS Settings BIOS settings differed between manufacturers, so we tuned for best-known SPECpower_ssj2008 performance results, following a similar strategy between the three systems. We disabled Turbo Mode on all blades which had it enabled by default and left Turbo off for blades that already had it disabled by default, and disabled all Processor Prefetcher Options exposed in BIOS, which generally improves java-oriented benchmarks such as SPECpower_ssj2008. For the Dell PowerEdge M610, we disabled Turbo Mode, and disabled the following Prefetcher options: Hardware Prefetcher, Adjacent Cache Line Prefetcher, DCU Prefetcher, and Data Reuse. We left the default Power Management mode (Dell Active Power Controller) intact. For the HP ProLiant BL460c G6, we changed the Dynamic Power Regulator speed on the HP blade to Slow (a common HP tuning to enhance performance/watt). We disabled Hardware Prefetcher and Adjacent Cache Line Prefetcher in BIOS. In addition, because two Intel Xeon processor 5600 series BIOS prefetcher settings (DCU Prefetch and Data Reuse) were not exposed as tunable options in the 2 From our testing, we determined that the Dynamic Power option in either position offered no real enhancement to SPECpower_ssj2008 results in the configuration as tested, so we left it at the default. Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 20 version of BIOS available at the time of our test, so we referred to the following HP site to disable these options to enhance performance using the conrep utility as described by HP Support Communication Customer Advisory Document ID: c02207408 version 23. For the IBM BladeCenter HS22, we disabled both exposed Processor Prefetcher options in BIOS: Cache Data Prefetch and Data Reuse. In addition, we enabled Processor C-States (disabled by default) to make the system more comparable to the HP and Dell blade solutions which have this option enabled by default. Enabling C-States enhances SPECpower_ssj2008 competitiveness by allowing the processor to reduce power to the cores to save energy at lower utilization levels, especially at idle. OS Tuning To improve Java performance, large pages were enabled by entering Control Panel->Administrative Tools->Local Security Policy->Local Policies->User Rights Assignment->Lock Pages in Memory. An option was changed to add Administrator. Operating System Power Management mode for all solutions was changed from Balanced (the default) to Power Saver and Power Saver mode was edited to turn off the Hard Drive after 1 minute. Since the IBM BladeCenter HS22 does not have a BIOS Power Management algorithm similar to Dell Active Power Controller or HP Dynamic Power Saver, we changed the minimum processor state of the Power Saver plan from 5% to 0% and the maximum processor state from 75% to 100%. This tuning mimics other Intel Xeon processor 5600 series based IBM SPECpower disclosures. We configured each blade with a separate IP address on the same subnet as our SPECpower_ssj2008 controller system where the Director, CCS, and PTDaemon components were located, and connected each blade to a Dell 24-port Gigabit switch external to the blade enclosure to ensure network connectivity to the controller system. SPECpower_ssj2008 Configuration IBM J9 Java Virtual Machine (JVM)4 was used for all three blade solutions, as this JVM provided the best performance for SPECpower_ssj2008 of any of the available choices at the time that this study was undertaken. The following JVM options were used on all three blade solutions, as they are the best-known JVM tunings for SPECpower_ssj2008 for the IBM J9 JVM when running with larger memory configurations: -Xmn1400m -Xms1875m -Xmx1875m -Xaggressive -Xcompressedrefs -Xgcpolicy:gencon -XlockReservation -Xnoloa Xlp The following bindings were used to ensure that each of the six JVMs ran on four logical processors: start /affinity [F,F0,F00,F000,F0000,F00000] 3http://bizsupport1.austin.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c02207408&lang=en&cc=us&taskId=101&prodSeriesId=3948598&prodTypeId=15351 4 JVM build 2.4, J2RE 1.6.0 IBM J9 2.4 Windows Server 2008 amd64-64 jvmwa64 60sr5-20090519_35743 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 21 Power Meter Configuration We used the Yokogawa WT210 Digital Power Meter for the actual power measurement of the blade enclosures, as this is the most commonly used analyzer for SPECpower_ssj2008 publications at the time that this study was undertaken. For the Dell blade solution, we used two Yokogawa WT210 Power Meters to measure the enclosure power by attaching the first three power supplies to one Power Distribution Unit (PDU) that was connected to a single WT210, and the next three power supplies to a second PDU connected to a second WT210. Each WT210/PDU combination was connected to a separate 208v floor outlet. See Figure 9. Figure 9. Power Measurement Diagram for the Dell Blade Solution PSU1 PSU2 PSU3 PSU4 PSU5 PSU6 PDU 208V 208V Dell M1000E Blade Enclosure Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 22 For the HP blade solution, we used two Yokogawa WT210 Digital Power Meters to measure the enclosure power by attaching the first three power supplies to one Power Distribution Unit (PDU) that was connected to a single WT210, and the next three power supplies to a second PDU connected to a second WT210. Each WT210/PDU combination was connected to a separate 208v floor outlet. See Figure 10. igure 10. Power Measurement Diagram for the HP Blade Solution The IBM blade solution presented a challenge. The IBM Blade Enclosure has two sets of three power cables coming from the rear of the unit. Because the power cable connector was different than the Dell and HP Blade Enclosures, we were forced to use a different PDU that would accept the NEMA C20 plug from the IBM enclosure. Each PDU only had two power receptacles to connect power cables, so we used three Yokogawa WT210 Digital Power Meters and three PDUs to measure power from this enclosure. Both of the cable bundles from the rear of the IBM enclosure contained three power cables. The first cable was marked PSU 1,2 , the next marked PSU 3,4 , and the last marked Blower. We attached the PSU 1,2 from the first cable bundle and PSU 1,2 cable from the second cable bundle to one Yokogawa WT210/PDU combination. The Blower power cables from both bundles were attached to a second Yokogawa WT210/PDU combination. Finally, the PSU 3,4 power cable from the first cable bundle and the PSU 3,4 power cable from the second cable bundle were attached to a third PSU1 PSU2 PSU3 PSU4 PSU5 PSU6 PDU 208V 208V HP c7000 Blade Enclosure Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 23 Yokogawa WT210/PDU combination. Each WT210/PDU combination was connected to a separate 208v floor outlet.5 See Figure 11. Figure 11. Power Measurement Diagram for the IBM Blade Solution 5 We also tried connecting power cable 1,2 and 3,4 from the first cable bundle to the first Yokogawa WT210/PDU combination, the blower cables to the second WT210/PDU combination, and power cables 1,2 and 3,4 from the second cable bundle to the third Yokogawa WT210/PDU combination, but got slightly lower SPECpower results using this cabling configuration, so we reverted to the above described cabling methodology (pairing similarly marked cables to PDUs). Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 24 Appendix B Enclosure Configuration Information Table 5. Enclosure Configuration Information Dell PowerEdge M1000e HP BladeSystem c7000 IBM BladeCenter H-Chassis 8852 Dimensions and Specifications Height (inches) 17.3 17.5 15.75 Width (inches) 17.6 17.5 17.5 Depth (inches) 29.7 32 29 U size in server rack 10 10 9 Number of blades 16 16 14 Chassis Management Firmware CMC 3.0.0 (build 32) Onboard Administrator 3.00 Advanced Module Management 3.54G Power Supplies Total number 6 6 4 Wattage of each 2700 2450 2900 Power Supply Part Number G803N 499243-B21 88524SU Cooling Fans Total number 9 10 2 Blower Modules I/O Modules in Chassis Gigabit Pass Through Modules 2 x Dell 16-port Gigabit Pass-Through 2 x HP 16-port Gigabit Pass-Through 2 x IBM 14-Port Gigabit Pass-Through Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 25 Appendix C Blade System Configuration Information Table 6. Blade System Configuration Information Servers Dell PowerEdge M610 HP ProLiant BL460c G6 IBM HS22 7870 4HU Memory Modules Total RAM in system (GB) 24 24 24 Vendor and model number Samsung M393B5170FHD-CH9 Micron MT36JSZF51272PY-1G4D1AB Samsung M392B5170EM1-CH9 Type PC3-10600R PC3-10600R PC3-10600R Speed (MHz) 1333 1333 1333 Speed in system as tested 1333 1333 1333 Timing/latency CAS 9 CAS 9 CAS 9 Number of RAM modules 6 x 4 GB 6 x 4 GB 6 x 4 GB Rank organization Dual Rank Dual Rank Dual Rank Hard Disk Vendor and model number Hitachi HUC151473CSS600 Seagate ST973452SS Seagate ST973452SS Number of disks in system 2 2 2 Size (GB) 73 73 73 Buffer size (MB) 16 16 16 RPM 15,000 15,000 15,000 Type SAS 6 Gbps SAS SAS 6 Gbps RAID Type RAID 1 RAID 1 + 0 RAID 1 Controller PERC H200 Modular SmartArray P410i LSI SAS StorPort 1064E Operating System Name Microsoft Windows Server 2008 R2 Enterprise Microsoft Windows Server 2008 R2 Enterprise Microsoft Windows Server 2008 R2 Enterprise Build number 7600 7600 7600 File system NTFS NTFS NTFS Language English English English Network Adapter Vendor and model number Broadcom BCM5709S NetXtreme II Broadcom 57711 10GbE Broadcom BCM5709S NetXtreme II Type Integrated Integrated Integrated Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 26 Appendix D Blade System Firmware and Drivers Table 7. Detailed Configuration Information for Blade Server Solutions Servers Dell PowerEdge M610 HP ProLiant BL460c G6 IBM HS22 7870 4HU Driver/Firmware Updates Network Firmware 5.0.13 2.1.5.7 2.1.3c Network Drivers 14.2.0 A00 5.2.20.0 5.2.17.0 HBA Firmware A01 2.74.0.0 2.70 HBA Drivers A01 6.20.0.64 1.30.04.00 HDD FW N/A N/A 1.0.6 Video Driver A02 6.14.10.6748 1.01.003 Power Management Controller Package N/A 3.4.0.0 N/A Management Controller Driver N/A 1.13.0.0 N/A Power Management Driver N/A 1.15.0.0 N/A Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 27 Appendix E SPECpower_ssj2008 Results This appendix provides actual detailed reports and results from Dell testing using tools from the Standard Performance Evaluation Corporation. Benchmark results stated above reflect results run as of July 6, 2010. For the latest SPECpower_ssj2008 benchmark results, visit http://www.spec.org/power_ssj2008/results/power_ssj2008.html. Dell M1000e Blade Enclosure/16 x PowerEdge M610 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 28 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 29 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 30 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 31 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 32 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 33 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 34 HP BladeSystem c7000/16 x ProLiant BL460c G6 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 35 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 36 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 37 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 38 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 39 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 40 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 41 IBM BladeCenter 8852 H-Series/14 x BladeCenter HS22 7870 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 42 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 43 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 44 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 45 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 46 Untitled DocumentPower Efficiency Comparison of Enterprise-Class Blade Servers and Enclosures 47
