IBM's blade servers are "fundamentally better" than HP's, according to IBM.
As part of the continuing war of words between the two server market leaders, IBM has claimed that, based on energy efficiency, its product outclass its rivals. Unsurprisingly, HP has questioned the credibility of IBM's claims, but not before fourth-placed Fujitsu Siemens had muscled into the battle, claiming that its product also uses significantly less energy than HP's.
Energy efficiency is a growing concern for data centre operators, so the attack is a well targeted effort to pull the booming market out of HP's hands.
IBM released test results which it claims show that its BladeCenter line of servers uses up to 30 per cent less electricity than HP's BladeSystem servers. The tests were done internally by IBM, however, raising doubts about them, although a company executive said Big Blue would welcome an independent analysis of both company's systems.
IBM compared its AMD Opteron-based BladeCenter server, model LS21 to an HP cClass BladeCenter, also Opteron-powered, model BL465c. The IBM blade was 30 per cent more energy efficient when idle and 18 per cent more efficient when running at full capacity. When comparing blades running a Xeon processor from Intel, the IBM HS21 blade is 26 per cent more energy efficient than the comparable HP model BL460c at idle and 13 per cent more energy efficient at full capacity, according to the tests. Both the Intel and AMD processors tested were dual core.
"Now we know why HP’s slogan is the 'lights out' data centre," began Doug Balog, vice president of IBM's BladeCenter product line. HP uses the marketing term "lights out" to boast that its data centres run efficiently around the clock without personnel to manage them, meaning they will run with the lights out. But Balog had a more waspish explanation: "As it turns out, turning out the lights offsets the inefficiency of their blade products."
HP responded with a prepared statement: "Without access to IBM's test methodology and results, it's hard to determine if valid, real-world scenarios were used. We are happy to be tested against IBM's best effort via an independent, unbiased and credible third-party."
Meanwhile, at its recent annual sales convention, Fujitsu Siemens Computers (FSC) was also running a comparison between its blades and HP blades. It claimed that in scenarios where HP blades used 2,000W, FSC ones used 1,750W. Scenarios included two-hour Exchange email sessions.
According to FSC, its enhanced blade server management allows it to decommission blades as demand falls. FSC CTO Dr Joseph Reger said: "We clearly out-perform - that is, use less energy - than HP." Reger also said that liquid cooling of blade servers was coming: "Liquid cooling is coming back in some shape or form." That's liquid cooling the entire rack, liquid cooling the chips, and/or liquid cooling entire data centres.
New power management tool
IBM also introduced an updated version of its PowerExecutive management tool for monitoring data centre energy use, which provides an overall view of energy usage in the data centre. The new version allows an operator to establish caps on energy usage and regulate the operation of servers to stay under that cap.
While Balog compared PowerExecutive version one to the gas mileage ratings for cars, he said version two acts like cruise control. But IDC analyst Vernon Turner questioned whether centre facilities managers should be able to control power usage if it conflicts with the role of IT specialists in charge of running certain applications on servers.
"It's very concerning that somebody may try to throttle the performance of a hardware device without knowing what's going on regarding the service level requirements for the applications," Turner said. "If somebody is controlling dials, so to speak, how do they know what the effect is on the overall infrastructure?"
IBM's Balog responded that IT managers and facilities managers collaborate more today than they may have in the past so there wouldn't be a conflict. But Turner also says blade servers are a small percentage of the server types in data centres today. And even if, according to industry estimates, blades may make up 30 percent of data centres by 2010, blade performance can't be considered in isolation. A centre has to look at its overall energy consumption versus its operational demands.
Original reporting by IDG News Service