This article is more than 1 year old

Cisco recalls suicidal UCS blade servers

Goodness gracious, great MOSFETs afire

Cisco Systems warns that its high-end B440 blades for its "California" Unified Computing System have a potentially disastrous defect that could result in one or more board failures, and emit a flash of light that could perhaps give system administrators heart attacks.

Last week, Cisco put out a field notice to customers using its UCS B440 server blades, saying that the failure of a MOSFET power transistor on the blade can "cause the component to overheat and emit a short flash which could lead to complete board failure." The company went on to say that "in extreme circumstances it could affect the other blades in the chassis by disrupting power flow."

Cisco actually warned customers that something was afoot with the MOSFETs back on July 12, 2011, and said at that time that there was "no indication of a systemic issue with the MOSFET components, and the observed failure in the field is considered to be a random component failure." To that end, Cisco's system engineers could whip up a firmware fix for the blade to keep the MOSFET from overheating and flashing, obviously causing the system board to fail. This is a no-no in the enterprise server racket.

Then on January 26, the company notified customers using the B440 servers that the firmware patch did detect MOSFET failures and prevent a "potential thermal event," but that since the firmware had been distributed, another B440 in the field has gone poof!. And so Cisco has made hardware modifications to the B440 system board and is now replacing all of the machines currently used by customers.

Cisco said in the field notice that no other UCS B Series blade servers or C Series rack servers are affected by this MOSFET failure issue.

If you have one of these B440s in production, Cisco recommends upgrading to the most recent UCS blade management controller software, which has the patch for monitoring the B440 MOSFETs, and arranging to get replacement blades as soon as possible.

The UCS B440-M1 blade server was launched back in April 2010, and is a double-wide blade that slides horizontally into the UCS chassis. It is a four-socket server that is based on Intel's eight-core "Nehalem-EX" Xeon 7500. The four-socket C460-M1 rack server has a slightly different design and is not affected by the MOSFET issue. Back in March 2011, Cisco updated its chubby UCS blade with the B440-M2, which is also a four-socket blade, but one supporting the ten-core "Westmere-EX" Xeon E7 processors from Intel. They shipped in last April.

There's no reason to believe that Cisco's distribution of servers is much different from the rest of the x86 world, so the MOSFET issue probably only affects somewhere between 5 and 10 per cent of its installed base. Cisco will very likely have more to say about the issue – and any costs it incurs for the recall – on Wednesday, when it goes over its financials for the second quarter of its fiscal 2012. ®

More about

More about

More about

TIP US OFF

Send us news


Other stories you might like