 | |||
|
|
||
| |||
| |||
| |
|
Home > Data Center News > IBM and HP spar over blade server superiority |
| |
|
|
|
|
| Data Center News: |
|
||
|
|||
|
|||
|
|||
|
|||
|
|||
|
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
SearchDataCenter.com was unable to locate HP blade users for this article with complaints about the overheating that creates memory depletion. But despite claims to the contrary, some users have reported heat issues in their data centers. Now, on the heels of a recent IBM study comparing IBM and HP blade servers, questions remain about whether these tests are useful and how users can judge which blade server is most appropriate for their data center environment. Results may beg questions, not answersJoe Clabby, an analyst at Yarmouth, Maine-based Clabby Analytics, agreed to sponsor IBM research comparing its blade servers with those of HP. This summer, Clabby visited an IBM BladeCenter benchmarking and testing laboratory in Raleigh, N.C., to oversee tests comparing the IBM BladeCenter H with the HP BladeSystem c-Class architecture and evaluate the results.
In conducting this isolated test, of course, IBM had an ulterior motive: that is, to demonstrate that under certain stress workloads, HP's blade memory modules run 10 degrees to 15 degrees hotter than the uppermost range typically recommended by most memory manufacturers, which is 90 degrees Fahrenheit. A similar test is also broadcast via YouTube by IBM's thermal engineer Vinod Kamath. Nevertheless, Clabby concurred with IBM's analysis. "HP blades run hotter than IBM BladeCenter under the same workload," Clabby reported. "What this means is that, in heavy workload environments, HP may be 'cooking' its memory modules [by] running memory out of spec for extended periods of time. ... This kind of situation can lead to some serious reliability/availability problems." In the test, HP equipment included 15 BL460x blades running dual 2.33 GHz Intel Xeon 5345 CPUs, with eight 2 GB dual in-line memory modules (DIMMS), one 73 GB 10k rpm disk drive, all in an HP c7000 enclosure with six power supplies and 10 fans. IBM's BladeCenter H included 14 HS21xm servers with the same Intel Xeon processors, 8X2GB DIMMs, four power supplies and two blowers. The test took place in a 77-degree room containing two pieces of test equipment: an Agilent Data Acquisition Switch Unit to measure voltage output, and an HP Testmobile Data Acquisition Module to measure temperature. In that same test environment, the IBM memory modules never exceeded the 85-degree mark, which is the high-end temperature range recommended by most manufacturers. The temp ran closer to 80 degrees throughout the entire test, Clabby reported. IBM's test followed tests publicized by HP in March that show HP BladeSystem c-Class uses up to 27% less power than the IBM BladeCenter-H in similar configurations in "real world" blade server environments. The Jim Ganthier, HP's director of blade systems division, categorically denies that overheating is an issue with c-Class blade servers and said this is "one more example of a contrived laboratory testing scenario by IBM." "Given long enough, you can make anything appear to fail," Ganthier said. "We have had no customers come back and tell us about this issue," Ganthier said. "We aren't seeing any data along these lines. We have measured our temps and seen only temps within the appropriate range." "Not that we have, but if we ever have a customer that sees high temperatures in worst-case conditions, we can right those issues. Blades were designed as an infrastructure platform, and we can cool each individual blade down with fans, but the Thermal Logic and onboard administrator would catch the issue before any problems ever occurred," Ganthier said. Ganthier said Clabby is an independent analyst who was paid for his IBM/HP analysis, insinuating partiality. But to be fair, HP's report from Sine Nomine Associates favorably comparing its blades with IBM's was paid for by HP. Is the proof in the users?As blade vendors battle it out under lab-testing conditions, users have their own experience to draw on with blades. And at least one HP c-Class user hasn't encountered problems with the blade chassis. Since early this year, Gentry Ganote, CIO of PGA Tour Superstore has been using HP c-Class blades and hasn't experienced overheating. He said his blade chassis resides in a colocation facility with top-of-the-line cooling. However, Martin MacLeod, a blade server consultant who writes the popular Blade Watch blog, said that some of his colleagues have had problems with disk failure, memory and CPU. Aaron Sawchuk, CTO of Quincy, Mass.-based ColoSpace, a data center hosting provider, said the first-generation blades from both vendors ran especially hot. "I have not seen that [memory burnout] scenario specifically, but I will say in general terms that the first generations of blades that both IBM and HP produced were fairly poorly designed with respect to airflow," Sawchuk said. "I do not think their engineers considered that their customers generally stack several enclosures into [a] single rack, thereby exacerbating the problem. Initially the negative results were attributed to poor rack design and airflow, but in reality it was a combination of both design and placement." Sawchuk went on to say, "The next generation of blades that have come out in the past year from IBM and more recently from HP have vastly improved their cooling design, yet ultimately the results for end users will depend on how where they place equipment in a data center and how they configure the cooling and heat evacuation for the individual cabinet or rack enclosures." Sifting through the agendaAnalysts say that given their marketing agenda, vendor-sponsored test results should be taken lightly. Illuminata Inc. analyst Gordon Haff has heard IBM's claims against HP's blades, and vice versa, at analyst conferences. He doesn't doubt the test results are true but questions whether the reported results have any meaningful correlation to actual reliability in real data center environments. MacLeod said he is always cautious about these kind of tests. "Let any blade get too hot, and the components will become less reliable and prone to failure," he emphasized. "Certainly we've had memory chip failures on the blade servers I've deployed and supported on IBM and HP blades. But not any more than you would expect based on the volume of blades deployed. Memory and disk seemed to be the weaker parts of the blade." Forrester Research Inc. principal analyst James Staten recommends that users take vendor-produced test results lightly. "Not that we have done an exhaustive study, but this is clearly an example of corner- and attack-type marketing. You can find vulnerabilities in anything and marketing can exploit it, but it is unlikely to occur in real life," Staten said. So with all the claims and counterclaims, whom should users believe? Well, neither. Staten suggests that users focus on which applications they run in the data center, their specific tools and the overall environment to make a sound decision on which blade to use. "If your data center ecosystem is HP oriented, you will adapt to HP's C-Class blades easily, and the same goes for people who are used to IBM servers. They will adapt quickly to IBM blades," Staten said. If you are new to blade servers, Staten suggests using the vendor whose servers you have used and liked in the past, because when you buy into blades and invest in a blade chassis, you are locking yourself into that vendor. Let us know what you think about the story; email Bridget Botelho, News Writer. Also, check out our news blog at serverspecs.blogs.techtarget.com.
|
|
|
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
 |  |  | |  | |  | |  | |  | |  | |  | |  | |
|
|
About Us | Contact Us | For Advertisers | For Business Partners | Site Index | RSS |
|
|
||
|
|||||||
