By Timothy Prickett Morgan
Gartner Group this week issued a warning to its clients alerting them to customer complaints about the reliability of Sun Microsystems Inc’s servers that employ UltraSparc-II processors in SMP configurations. In particular, says Gartner, the problem seems to affect Sun servers that are configured with 400MHz processors with 4Mb of static RAM L2 cache memory. The 400MHz UltraSparc-IIs equipped with 8Mb of L2 cache do not seem to be affected.
Garter says it has heard customer complaints from companies using more than 20 processors in a UE6500 server and more than 36 processors in a UE10000 Starfire server. Moreover, the consultancy says that customers with lighter configurations have also had troubles. Sun has acknowledged to Gartner that it did have an SRAM problem with the 4Mb L2 cache units, and has said that it had troubles in its fiber optic I/O controllers for the servers as well. While it said the problems were fixed months ago, Gartner’s report cited problems that occurred only a few weeks back.
Gartner says it’s not a problem of machines crashing necessarily, but is concerned that Sun’s Starfire accounts are seeing availability drop to 94% to 98%; a far cry from the 99.95% availability promise that Sun sells the Starfire with. The best approach to the problem is to rip all of the UltraSparcs with 4Mb L2 cache out of the Sun machines and replace them with chips that use 8Mb L2 caches. Obviously, customers buying new Sun iron should not accept bids on 4Mb L2 cache units until Sun has cleared up the problem once and for all. It is always a good idea to get replacement part contingency plans in writing from the start, too.
