Guest

Power Efficiency Comparison: Cisco UCS 5108 Blade Server Chassis and HP BladeSystem c7000 Enclosure

  • Viewing Options

  • PDF (2.3 MB)
  • Feedback

Contents

Executive Summary

Main Findings

Test Method Overview

Hardware Configuration 1

Results: Hardware Configuration 1

Hardware Configuration 2

Results: Hardware Configuration 2

Conclusion

Appendix A: Solution Firmware and Driver Details

Appendix B: Test Procedure

Hardware and System Firmware

BIOS

Operating System

Benchmark

Java Virtual Machine

Power and Temperature Measurements

Appendix C: SPECpower_ssj2008 Test Results


Executive Summary

Rising energy prices and a drastic increase in computing solution density have made energy management a critical component of efficient data center operations. IT managers must take into account power utilization and planning, in addition to, their data center availability, scalability, and resource utilization requirements to cost-effectively deliver solutions for their business. Cisco is dedicated to the advancement of energy efficiency in the data center and in computing ecosystems. This document compares the power efficiency of similarly configured Cisco UCS ® 5108 Blade Server Chassis and HP BladeSystem c7000 enclosures.
Cisco compared the power and performance characteristics of equivalently configured Cisco UCS 5108 Blade Server Chassis and HP BladeSystem c7000 enclosures. The Cisco system was configured with the Cisco UCS B200 M3 Blade Server, and the HP system was configured with the equivalent HP ProLiant BL460c Gen8 Server Blade. Both systems used the Intel Xeon processor E5-2600 product family. SPECpower_ssj2008, was run on both configured solutions to evaluate performance, power consumption, and power efficiency.
The results demonstrated that the Cisco UCS 5108 enclosure configured with Cisco UCS B200 M3 blades was more power efficient and consumed less power than the HP BladeSystem c7000 configured with HP ProLiant BL460c Gen8 blades.

Main Findings

• Efficiency

– The Cisco UCS 5108 enclosure configured with Cisco UCS B200 M3 blades achieved a 6.3 percent higher performance-to-power ratio than the HP BladeSystem c7000 with HP ProLiant BL460c blades, using with similarly configured hardware, BIOS, and OS settings.

– At a 70 percent target load, the Cisco UCS blade solution consumed 6.5 percent less power than the HP ProLiant blade solution.

• Power

– At the maximum target load, the Cisco UCS blade solution consumed 192 watts (W), or 7.2 percent less power, than the equivalently configured HP BladeSystem solution.

– The Cisco UCS blade solution consumed 167 watts, or 17.4 percent less power, while idle than the equivalently configured HP BladeSystem solution: a difference equivalent to the power consumed by three idle HP Proliant BL460c Gen8 blades.

– Including the power consumed by a pair of redundant fabric interconnects, the Cisco UCS blade solution consumes less power than an equivalent HP BladeSystem solution (see Figure 6 later in this document).

Test Method Overview

To make relevant power consumption and efficiency comparisons, all solution variables that affect power consumption and performance must be equivalent. The Cisco UCS 5108 and HP BladeSystem c7000 blade solutions were similarly configured. The BIOS parameters affecting performance and power consumption were optimized and set equivalently for each blade solution (see Appendix B for detailed BIOS parameter configuration information). The same benchmark workload was run on each blade solution operating in the same environmental conditions.
A benchmark provides a basis for comparison between equivalent computing solution offerings. SPECpower_ssj2008 is a benchmark developed by the Standard Performance Evaluation Corporation (SPEC), a nonprofit group of computer vendors, system integrators, universities, research organizations, publishers, and consultants. The benchmark is designed to provide a view of server system power consumption while the system runs Java server applications. The results from each SPECpower_ssj2008 benchmark are listed in Appendix C. These results provide the data that supports the analysis in this report.

Hardware Configuration 1

The first blade solution evaluated was a single enclosure configured with eight blades. The Cisco and HP blade enclosures were configured similarly, with comparable management hardware components (Table 1). To measure power consumption during benchmark execution, a power analyzer was connected to the power supply input of each blade enclosure.

Table 1. Blade Solution (Configuration 1)

Enclosure

Cisco UCS 5108

HP BladeSystem c7000

Blade Slots Available/Installed

8/8

16/8

Enclosure Management Module

UCS 6248UP [2]1

Onboard Administrator Module [2]

Internal IO Module

UCS 2204XP Fabric Extender [2]

VC FlexFabric 10Gb Module [2]

Power Supply

2500W Platinum Rated [4]

2400W Platinum Rated [4]

Fan Slots Available/Installed

8/8

10/8

Blade Model

B200 M3

BL460c Gen8

Form Factor

Half Height

Half Height

Processor

Intel Xeon E5-2660 [2]

Intel Xeon E5-2660 [2]

Physical/Logical Cores

16/32

16/32

Memory

4GB DDR3 RDIMM PC3L-12800 [8]

4GB DDR3 RDIMM PC3L-12800 [8]

Hard Disk Drive

300GB 10K 6Gb, RAID 0 [1]

300GB 10K 6Gb, RAID 0 [1]

Network

VIC 1240 10Gb 4-port Adapter [1]

FlexFabric 10Gb 2-port Adapter [1]

Storage Controller

LSI Logic SAS 2004 [1]

Smart Array P220i [1]

1. The Cisco UCS 6248UP 48-Port Fabric Interconnect provides the management and communication backbone for the Cisco UCS B-Series Blade Servers and C-Series Rack Servers. A single redundant pair of fabric interconnects supports up to 20 blade enclosures in a single highly available management domain. In the power efficiency comparison for this configuration, the power consumed by the redundant pair of fabric interconnects was not included in the measurement.
The latest available system firmware at the time of testing was downloaded and installed; see Appendix A for additional details.
Microsoft Windows Server 2008 R2 Enterprise with Service Pack 1 (SP1) was installed on each server. After installation, the update utility was run, and all available updates at the time of testing were installed. The same operating system power management settings were used for both solutions; see the Operating System section of the test procedure in Appendix B for additional details.
Using the Cisco UCS service profile feature, all Cisco UCS blade BIOS parameters were updated simultaneously using a common BIOS policy. Each BIOS parameter was manually duplicated on the HP blade servers using the HP ROM-Based Setup Utility; see the BIOS section of the test procedure in Appendix B.
The blade enclosure power management policy was set to grid redundant for each solution. Only the enclosure power supplies were connected to the power meter. External networking components were not included in the power measurements.

Results: Hardware Configuration 1

At maximum target load, the Cisco UCS blade enclosure consumed 192W, or 7.2 percent less power than the equivalent HP blade solution. In the active-idle state, the Cisco UCS blade enclosure consumed 167W, or 17.4 percent less power than the HP blade solution.
The Cisco UCS blade solution consumed less power while providing performance comparable to that of the HP blade solution. Thus, the Cisco UCS blade solution was more efficient, with a 6.3 percent greater performance-to-power ratio than the HP blade solution (Figure 1).

Figure 1. Blade Solution Power Efficiency (Configuration 1)

The benchmark computes power efficiency by dividing the number of Java server application operations performed by the average power consumed at each target interval. The performance per watt for each interval is shown in Figure 2. At each target load, the Cisco blade solution demonstrated a power efficiency advantage over the comparable HP blade solution. The Cisco blade solution advantage was 4.1 to 8.0 percent better performance per watt across all target loads.

Figure 2. Target Load Power Efficiency (Configuration 1)

The methods used to estimate and plan power requirements vary widely. A general rule used by many IT professional is to size the solution based up power consumption at a specified target workload. The typical target workloads range from 50 to 70 percent utilization. For this comparison, power consumption at a 70 percent SPECpower_ssj2008 target load was plotted as shown in Figure 3. At this reference point, the Cisco UCS blade solution consumes 127W, or 6.5 percent less power than the comparable HP blade solution.
Another efficiency metric used by IT professionals is the amount of power consumed while the system is idle. The SPECpower_ssj2008 benchmark includes an active-idle measurement. The average power consumed by the Cisco blade solution was 167W, or 17.4 percent less than the equivalent HP blade solution, as shown in Figure 4.

Figure 3. Blade Solution Power Efficiency at 70 Percent Target Load (Configuration 1)

SPEC Fair Use Rule disclosure condition: At 70 percent target load Cisco UCS B200 M3 Blade Server achieved 7,902,138 ssj_ops using 1824W, and HP ProLiant BL460c Gen8 blade solution achieved 8,119,073 ssj_ops using 1951W.

Figure 4. Blade Solution Active-Idle Power (Configuration 1)

Hardware Configuration 2

Blade solutions can be extremely efficient. For that reason, enterprise customers deploy multiple blade enclosures, to achieve the best power density and gain management advantages compared to traditional rack servers.
Configuration of a large-scale hardware solution, consisting of hundreds of blade servers, to measure power efficiency is not practical. Instead, a base hardware unit is used to scale and extrapolate power efficiency. The logical base unit is a fully configured blade enclosure. The HP BladeSystem c7000 supports 16 blades per enclosure. Thus, a 16-blade solution was used as the base hardware unit (Table 2). The base hardware unit is duplicated for each 16-blade increment.

Table 2. Base Hardware Unit (Configuration 2)

Enclosure

2X Cisco UCS 51081

1X HP BladeSystem c7000

Blade Slots Available/Installed

2X 8/8

16/16

Enclosure Management Module

UCS 6248UP [2]2

Onboard Administrator Module [2]

Internal IO Module

2X UCS 2204XP Fabric Extender [2]

VC FlexFabric 10Gb Module [2]

Power Supply

2X 2500W Platinum Rated [4]

2400W Platinum Rated [6]

Fan Slots Available/Installed

2X 8/8

10/10

Blade Model

B200 M3

BL460c Gen8

Form Factor

Half Height

Half Height

Processor

Intel Xeon E5-2660 [2]

Intel Xeon E5-2660 [2]

Physical/Logical Cores

16/32

16/32

Memory

8GB DDR3 RDIMM PC3L-10600 [4]

8GB DDR3 RDIMM PC3L-10600 [4]

Hard Disk Drive

300GB 10K 6Gb, RAID 0 [1]

300GB 10K 6Gb, RAID 0 [1]

Network

VIC 1240 10Gb 4-port Adapter [1]

FlexFabric 10Gb 2-port Adapter [1]

Storage Controller

LSI Logic SAS 2004 [1]

Smart Array P220i [1]

1. Two Cisco UCS 5108 blade enclosures are required to achieve the base hardware unit.
2. The Cisco UCS 6248UP 48-Port Fabric Interconnect is a core part of Cisco UCS. Typically deployed in redundant pairs, the Cisco UCS 6248UP provides uniform access to both network and storage. The Cisco UCS fabric extender architecture provides management for 20 blade enclosures in a single unified system without additional complexity, thus eliminating dedicated chassis management and blade switches and reducing the number of cables required.
The procedures from the previous analyses were used to measure performance and power consumption with one exception: the power consumed by the pair of redundant Cisco UCS 6248UP fabric interconnects was included during execution of the SPECpower_ssj2008 benchmark.

Results: Hardware Configuration 2

Beginning with the base hardware unit, the total power consumption was compared for each 16-blade increment. A single pair of fabric interconnects can support up to 20 blade enclosures, or 160 blades, so the power consumed by the redundant pair of interconnects was added once to each 16-blade increment for comparison. For example, the power consumed by an 80-blade solution was calculated using 10 Cisco UCS 5108 enclosures and 2 Cisco UCS 6248UP fabric interconnects, and using 5 HP BladeSystem c7000 enclosures (Figure 5). A single Cisco UCS enclosure with 8 blades consumed 1991W. With the same target load, the power consumed by the redundant pair of fabric interconnects was 530W (see Appendix C). With the same target load, a single HP blade enclosure consumed 4378W. The calculations for projecting power consumption for an 80-blade solution were 10 x 1991W + 2 x 265W, or approximately 20.4 kW for the Cisco UCS solution; and 5 x 4378W, or approximately 21.8 kW for the HP BladeSystem solution.

Figure 5. Example of Total Power Comparison Using an 80-Blade Solution

With the first increment, the Cisco UCS solution consumes 129W more power than the equivalent HP BladeSystem solution. With the next increment, the Cisco solution draws 267W less power than the equivalent HP BladeSystem solution. The unified fabric power savings advantage becomes more apparent as the number of blade enclosures increases, because for every HP blade enclosure added to the system, duplicate onboard administrators and virtual connect switches are needed. As the number of blades increase, the power savings decisively favors the Cisco UCS blade solution. The projected power savings at 160 blades was greater than 3.4 kW; see Figure 6 and Figure 7, which show the power savings across the entire range.
Using 160 Cisco blades instead of HP ProLiant blades, the end user can save more than 3.4 kW of power. The power savings is nearly equivalent to turning off an HP c7000 BladeSystem fully configured with HP ProLiant blades. The projected power savings is a conservative estimate that does not include the additional power savings from data center infrastructure. Additional infrastructure power is required for distribution (transformer and conversion losses) and cooling (mechanical cooling). Assuming a facility power usage efficiency (PUE) rate of 1.6, the projected total power savings is 5.5 kW.

Figure 6. Projected Cisco UCS Power Savings

Figure 7. Projected Power per Blade

SPEC Fair Use Rule disclosure condition: At 90 percent target load Cisco UCS B200 M3 Blade Server achieved 9,265,289 ssj_ops using 1991W, and HP ProLiant BL460c Gen8 blade solution achieved 20,208,798 ssj_ops using 4378W.

Conclusion

The SPECpower_ssj2008 benchmark results yielded comparable performance for both blade solutions. Both blade solutions were configured with similar hardware and firmware running the same workload; however, the Cisco blade solution consumed less power from the active-idle state through the maximum target load.
The first configuration measured the power efficiency of the blade, blade enclosure, and internal management components of a typical high-availability deployment. The Cisco UCS solution with Cisco UCS B200 M3 blades had 6.3 percent greater power efficiency than the HP BladeSystem solution with HP ProLiant BL460c Gen8 blades. With a 70 percent target load, the Cisco UCS solution used 6.5 percent less power than the HP BladeSystem solution. In the active-idle state, the Cisco UCS solution consumed 17.4 percent less power than the equivalent HP BladeSystem solution.
The second configuration provided a construct for comparing power efficiency through scaling. The comparison used a base hardware unit of measure: two Cisco UCS 5108 blade enclosures with a redundant pair of Cisco UCS 6248UP fabric interconnects compared to one HP BladeSystem c7000 blade enclosure. Even including the power consumed by the pair of redundant fabric interconnects, the Cisco UCS blade solution consumed less power than an equivalent HP BladeSystem solution. The Cisco solution used 267W less power than the equivalent 32-blade HP BladeSystem solution. As the number of blades increased, the power savings steadily increased in favor of the Cisco UCS blade solution. The projected power savings at 160 blades was more than 3.4 kW.
Over the years, computing solutions have become less expensive to purchase and maintain, delivering more computing capacity at lower equipment and operating costs. At the same time, the cost of energy has continued to increase. The industry is at the point where the cost of electricity to run and cool computing systems can exceed the cost of the initial system purchase. With its higher performance-per-watt power ratio, the Cisco UCS solution provides a more efficient platform than the HP BladeSystem solution.

Appendix A: Solution Firmware and Driver Details

Table 3 provides firmware and driver details for the Cisco UCS 5108 Blade Server and HP BladeSystem c7000 solutions.

Table 3. Enclosure and Blade Solution Firmware and Driver Configurations

Component

Cisco

Hewlett Packard

Enclosure Model

UCS 5108

Blade System c7000

Enclosure Management Firmware

5.0(3)N2(2.04b)

3.70

Internal IO Module Firmware

2.0(4b)

3.70

Blade Model

B200 M3

BL 460c Gen8

System BIOS

B200M3.2.0.4a.0.080920121557

I31 08/20/2012

Integrated Management Controller Firmware

2.0(4b)

1.10

Network Adapter Firmware

2.0(4b)

4.1.450.5

Storage Controller Firmware

20.10.1-0061

3.22

Power Mangement Controller Firware

N/A

3.0

Management Controller Driver

N/A

3.4.0.0

Network Adapter Driver

2.1.0.10

4.0.449.0

Storage Controller Driver

5.1.112.64

6.22.0.64

Display Adapter Driver

6.1.7600.16385

1.2.1.0

Appendix B: Test Procedure

This appendix describes the test procedure used to compare the Cisco UCS 5108 Blade Server and HP BladeSystem c7000 solutions.

Hardware and System Firmware

Each blade solution was configured similarly with comparable hardware components; see Tables 1 and 2 earlier in this document for hardware configuration details. The latest available system firmware at the time of testing was downloaded and installed; see Appendix A for details.

BIOS

The available BIOS parameters differ between the Cisco and HP blade solutions. The BIOS parameters were set as equivalently as possible to help ensure comparable performance and power efficiency; see Table 4 for the BIOS settings configured for each blade solution.

Table 4. Blade Solution BIOS Settings

Component

B200 M3

BL460c Gen8

Processor Configuration

Intel HyperThreading Technology

Enabled

Enabled

Number of Enabled Cores

All

All Cores Enabled

Intel VT

Disabled

Disabled

Intel VT-d

Disabled

Disabled

Processor Performance Configuration

CPU Performance

Enterprise1

-

Workload Configuration2

Balanced

Balanced Performance

Hardware Prefetcher

Disabled

Disabled

Adjacent Sector Prefetcher

Disabled

Disabled

DCU Streamer Prefetcher

Disabled

Disabled

Processor Power Management Configuration

Enhanded SpeedStep Technology

Enabled

Enabled

Intel Turbo Boost Technology

Enabled

Enabled

Processor Power State C1 Enhanced

Enabled

-

Processor Power State C6

Enabled

Enabled

Energy Performance Policy3

OS Controlled

OS Controlled

HP Power Profile4

-

Balanced Performance

Memory Configuration

Operating Memory Voltage

Low Voltage

Optimized for Power5

Low Voltage DDR Mode

Power Savings Mode

Balanced6

QPI Configuration

QPI Link Frequency Select

6.4GT/s

Min QPI Speed7

USB Configuration8

USB Control

-

USB Enabled

USB Boot Support

-

Enabled

Removeable Flash Media Boot Sequence

-

External DriveKey First

1. Selecting Enterprise disables all prefetchers and data reuse: for example, the hardware prefetcher, adjacent cache line prefetcher, and data cache unit (DCU) prefetcher.
2. The equivalent HP BIOS parameter is HP Power Regulator, which uses the operating system power management policy. The operating system used, Microsoft Window Server 2008, ignores this parameter regardless of its setting.
3. The equivalent HP BIOS parameter is Energy/Performance Bias. Balanced Performance is the default setting; it provides the best balance of power efficiency and performance and is recommended by HP for most environments.
4. Although the parameter was set to Balanced Power and Performance, it is automatically set to Custom when individual power settings are modified, as in this case.
5. The equivalent HP BIOS parameter is DIMM Voltage Preference. Optimized for Power Low Voltage DIMMs run at 1.35V, even if this voltage requires memory to run at a lower frequency.
6. The equivalent HP BIOS parameter is Memory Power Savings Mode. Balanced is the default setting for optimal power efficiency.
7. The minimum Intel QuickPath Interconnect (QPI) Link Frequency rate supported by the Intel Xeon processor E5-2600 product family is 6.4 gigatransfers per second (GTps).
8. All USB ports are enabled by default on the Cisco UCS B200 M3.
For a complete list of available BIOS settings for the Cisco UCS B200 M3 and HP BL460c Gen8, see the following links:

Cisco UCS Manager GUI Configuration Guide, Release 2.0 - Configuring BIOS Settings

HP ROM-Based Setup Utility User Guide (HP Part Number: 347563-406)

Operating System

The same Microsoft Windows Server 2008 R2 Enterprise with Service Pack 1 image was installed on each blade server. After the operating system was installed, the update utility was used to install all recommended and optional updates available at the time of testing.
The same operating system power management settings were used on both blade solutions. The power management plan was set to Balanced. The specific settings are shown in Figure 8.

Figure 8. Operating System Power Management Options

To improve Java performance, the local security policy was modified to enable the Administrator account to lock pages in memory. The security setting determines which accounts can use a process to keep data in physical memory, which prevents the system from paging the data to virtual memory on disk.

Benchmark

The latest SPECpower_ssj2008 version, Version 1.12, was installed on each blade and on the control system. The SPECpower_ssj2008 benchmark was developed by the Standard Performance Evaluation Corporation (SPEC), a nonprofit group of computer vendors, system integrators, universities, research organizations, publishers, and consultants. It was designed to provide a view of a server system's power consumption running Java server applications.
SPECpower_ssj2008 consists of three main software components:

• Server-Side Java (SSJ)-Workload

– SSJ-Workload is a Java program designed to exercise the CPUs, caches, memory, scalability of shared-memory processors, Java Virtual Machine (JVM) implementations, just-in-time (JIT) compilers, garbage collection, and other aspects of the operating system of the system under test (SUT).

– For more information about SSJ, see http://www.spec.org/power/docs/SPECpower_ssj2008-Design_ssj.pdf.

• Power and Temperature Daemon (PTDaemon)

– PTDaemon offloads the work of controlling a power analyzer or temperature sensor during measurement intervals to a system other than the SUT.

– For more information about PTDaemon, see http://www.spec.org/power/docs/SPEC-PTDaemon_Design.pdf.

• Control and Collect System (CCS), including Visual Activity Monitor (VAM)

– CCS is a multithreaded Java application that controls and enables the coordinated collection of data from multiple data sources such as a workload running on a separate SUT, a power analyzer, and a temperature sensor.

– VAM is a software package designed to display activity from one, two, or three SUTs simultaneously in combination with the SPECpower_ssj2008 benchmark.

– For more information about CCS and VAM, see http://www.spec.org/power/docs/SPECpower_ssj2008-Design_ccs.pdf.

All results discussed in this document are from compliant runs. Although the tests have not been submitted to SPEC for review, Cisco can disclose the results for the purpose of this study. The comparisons comply with the required conditions outlined in the SPEC Fair Use Rule for SPECpower_ssj2008. All details required to reproduce these results are listed in the appendixes. The files from each complaint run referenced in this document are included in Appendix C.

Java Virtual Machine

The same JVM version was installed on each blade and the control system. The JVM version installed was IBM J9 Virtual Machine (VM): Build 2.6, Java Run Environment (JRE) 1.7.0, Microsoft Windows Server 2008 R2 amd64-64 20120322_106209, JIT enabled, and ahead-of-time (AOT) compilation enabled.
The same JVM command-line options were used on all blades:
-Xgcpolicy:gencon -Xaggressive -Xcompressedrefs -Xmx1500m -Xms1500m -Xmn1100m
-XlockReservation -Xnoloa -XtlhPrefetch -Xlp -Xconcurrentlevel0
A complete list of JVM command-line options and their functions can be found in the IBM user guides for Java V7 on Microsoft Windows.
Each blade was configured with two Intel Xeon processors E5-2660 CPUs, eight cores per socket, and two threads (logical processors) per core. Eight JVM instances were started on each blade. Each JVM instance was bound to four logical processors. The following CPU affinity commands were used:
start /high /affinity [F,F0,F00,F000,F0000,F00000,F000000,F0000000]

Power and Temperature Measurements

Yokogawa WT210 Digital Power Meters were used to collect power measurements. The Yokogawa WT210 units used were within calibration limits.
The enclosures were mounted in the same rack. The inlet temperature was measured at the front of each blade enclosure during testing. A Digi International Watchport/H probe was used to collect temperature and humidity data. The inlet temperatures recorded were approximately 24.5 °C and varied less than 0.5 °C between each test.

Appendix C: SPECpower_ssj2008 Test Results

Figures 9 through 13 provide SPECpower_ssj2008 test results for the Cisco UCS 5108 Blade Server and HP BladeSystem c7000 solutions.

Figure 9. SPECpower_ssj2008 Results for Cisco UCS B200 M3 (Configuration 1)

Figure 10. SPECpower_ssj2008 Results for HP BL460c Gen8 (Configuration 1)

Figure 11. SPECpower_ssj2008 Results for Cisco UCS B200 M3 (Configuration 2)

Figure 12. SPECpower_ssj2008 Results for Cisco UCS B200 M3 with Fabric Interconnects (Configuration 2)

Figure 13. SPECpower_ssj2008 Results for HP BL460c Gen8 (Configuration 2)