The Real Story about IBM’s BladeCenter power claims

December 2006 — IBM is aggressively attacking HP’s new BladeSystem c-Class.  There’s a lot of confusing and contradictory data about blade energy efficiency out there – it’s hard to know what to compare and how.  Recently, IBM claimed that their blades were significantly more energy efficient than HP.¹   

To see which of their claims to energy efficiency could stand up to scrutiny, HP put IBM’s BladeCenter and BladeCenter H into a head-to-head test, under equal comparison, versus the HP BladeSystem c-Class.  This is what we found.

Fact 1: HP attempted to recreate a configuration where the IBM BladeCenter H demonstrated a power efficiency advantage over the HP BladeSystem c7000 -- but could find none.[2]

Fact 2:  Using the older IBM BladeCenter the testing team could identify just one corner case where IBM demonstrated a power efficiency advantage over the HP BladeSystem c7000.  This corner case was an IBM BladeCenter (not H) with just ONE (1!) HS21 blade server versus an HP BladeSystem c7000 with just ONE BL460c, both running at idle.[3] 

Fact 3:  HP testing showed that a fully configured HP BladeSystem c7000 used less energy per blade than a fully configured IBM BladeCenter H both at idle and fully stressed conditions.[7] 

Fact 4:  HP testing showed that the IBM BladeCenter H enclosure requires more air flow[8]  than the HP BladeSystem c7000 enclosure.

Fact 5: HP testing showed that the IBM BladeCenter H produced more noise[11] than the HP BladeSystem c7000.

Fact 6: HP testing showed that IBM Power Executive incorrectly reported power usage by as much as 42%, while the HP Onboard Administrator power monitoring accuracies were within 1% of independent power meter readings.

Adding to the inaccuracy of IBM Power Executive, HP testing found that IBM Power Executive did not measure the power drawn by their blade enclosure fans.  HP engineers found that the IBM BladeCenter H fans are on a separate AC circuit, and there is no path for the management tool to capture or monitor blade enclosure fan power.¹²  You can’t manage what you can’t measure.

To accurately manage the power and cooling in the datacenter customers have to be able to rely on accurate tools, Like HP’s Onboard Administrator.

Let’s discuss 2 more considerations above and beyond the blade system.

Fact 7: IBM states that a fully populated 42U rack of IBM BladeCenter enclosures requires 30kW of power.[14]  However, the IBM Rear Door Heat eXchanger is specified to remove only 15kW of heat.[15]  

Fact 8: HP Dynamic Smart Cooling is the industry’s first solution that integrates datacenter-wide sensors, intelligent control, and automated provisioning that allows the computing infrastructure to interact with the datacenter facilities to optimize cooling.[17]

¹ “IBM’s claims and methods are summarized in the November 16, 2006 press release
IBM BladeCenter Systems Up to 30 Percent More Energy Efficient Than Comparable HP Blades;
http://www-03.ibm.com/press/us/en/pressrelease/20633.wss

²  HP engineering testing

³  HP engineering testing

⁴  Tested configurations: IBM BladeCenter H with  14 – IBM HS21 server blades  with 2 – 2.33GHz Intel 5140 processors, 4 – 1GB DIMM configured in single branch , 2 – 36GB, 10K disk drives; RAID 0, Enclosure:  2 – blowers, 4-power supplies, 2 – IBM Ethernet switches, 1-management module.
HP BladeSystem c-Class:  16 – HP ProLiant BL460c server blades with 2 – 2.33GHz Intel 5140 processors, 4 – 1GB DIMM configured in single branch , 2 – 36GB, 10K Hot-plug disk drives; RAID 0.  Enclosure:  10- active cool fans, 4 –power supplies, 2 – blade network technology switches, 1- Onboard Administrator module.

⁵  Idle is “OS idle” under Windows2003® unless specified otherwise.

⁶  Stress, using the Prime95 FFT Torture test found at www.mersenneforum.org, unless specified otherwise.

⁷  IBM BladeCenterH:  idle power 188 volt-amps per blade, fully loaded power 277 volt-amps per blade.
HP BladeSystem c-Class:  idle power 176 volt-amps per blade, fully loaded power 269 volt-amps per blade.
(Volt-amps measured using a Voltech PM300 power meter with shunt connection on each phase)

⁸  Airflow measurements were made using ANSI/AMCA Standard 610-95, "Methods of Testing Airflow Measurement Stations for Rating"

⁹ IBM BladeCenterH:  idle air flow 31.9 CFM, fully loaded air flow:  31.9 CFM.
HP BladeSystem c-Class:  idle air flow 16.5 CFM, fully loaded air flow: 19.1 CFM
CFM – cubic feet per minute

¹⁰  HP, Christopher Malone, PhD, Christian Belady, P.E., “Metrics to Characterize Data Center & IT Equipment Energy Use”, Digital Power Forum, Richardson, TX (September 2006) & “How to Minimize Data Center Utility Bills”, HP C. Belady, P.E., Sept 2006, https://www.line56.com/articles/default.asp?ArticleID=7881&pg=1&topicID=0

¹¹ Acoustic testing was completed per ISO 9296 standards, in the HP Acoustics Laboratory; an ISO/IEC 17025:2005 accredited test facility.

¹²  IBM BladeCenterH:  fully loaded  7.8 Bels , fully loaded noise  with a fan failure 9.3 Bels
HP BladeSystem c-Class fully loaded  7.4 Bels, , fully loaded noise  with a fan failure 7.4 Bels
Acoustics LWAD (Bels) (lower is better).  In this case the stress test was accomplished with an internal HP stress test, MG.

¹³ HP engineering analysis

¹⁴ “BladeCenter packaging, power, and cooling” http://researchweb.watson.ibm.com/journal/rd/496/crippen.html

¹⁵  IBM says “Up to 50,000 BTUs of heat removed (or approximately 15 kW) from air exiting the back of a rack full of servers” Source:
http://www-132.ibm.com/webapp/wcs/stores/servlet/ProductDisplay?catalogId=-840&storeId=1&langId=-1&dualCurrId=73&categoryId=4611686018425028106&productId=4611686018425023461

¹⁶  http://h18004.www1.hp.com/products/servers/proliantstorage/racks/mcs/index.html

¹⁷ Based on the analysis of current and comparable offerings in the industry

¹⁸  ZDNET Blog “Between the Lines”,  http://blogs.zdnet.com/BTL/?p=4026 writes:
“[IBM] claims to offer similar capabilities to HP's forthcoming Dynamic Smart Cooling. However, I didn't find a similar product or technology concept in IBM's Cool Blue portfolio of datacenter energy savers. Earlier this month IBM claimed that its BladeCenter system uses up to 30 percent less energy than HP's comparable BladeSystem, touting energy efficient power supplies and high-efficiency blowers for blade racks, and the second generation of PowerExecutive, IBM's energy management technology that allows the amount of power used by a single server or groups of servers to be capped.  But nothing about dynamic control of CRAC systems. I'm waiting to hear back from IBM to clear up this mystery…or bit of dissembling by IBM.