Hello Simon, The opteron system we tested was a 1U 2cpu system from penguin. Despite the turmoil of the previous thread, it passed all our tests and has performed hansomely in benchmarks. The problems discussed in the thread were tracked down to (1) a bad ram chip and (2) obscure errors with the ncbi toolkit that had nothing to do with opteron. IBM seems ready to go with opteron clusters, and their technical staff was very helpful in tracking down some of our issues. Networx is installing some *big* clusters these days - haven't had any personal experience with them. Take care, Nathan Vsevolod (Simon) Ilyushchenko wrote: > Hi all, > > We are looking at buying a fairly large cluster (several hundred nodes) > next year. If you have any thoughts on the following questions, could > you please share them? > > 1. What are your top three cluster vendors in terms of hardware > reliability and support? > > 2. Has anyone used IBM/HP/Sun blades in Linux clusters? Are there any > blade-specific problems? (Eg, overheating, bottlenecks in chassis > throughput...) > > 3. Is anyone using (happily) a half-U non-blade solution? (Meaning > two dual-CPU machines in a 1-U space.) If yes, who is the vendor? > > 4. What is your favorite method of file sharing for hundreds of nodes? I > tested PVFS over a year ago, but it was slow and unreliable. Now there > also are OpenGFS and Lustre... Are there any hardware solutions for this > which don't cost insane amounts of money (like SAN connection on every > node)? > > 5. I saw the recent thread about problems with Opterons. How close in > your opinion are they to production use? How about Itaniums? I see on > the web a lot of press releases about Itanium clusters, but very few > real-life stories. Has anyone heard of 64-bit CPUs on blades in the > pipeline? > > Thanks a lot, > Simon -- Nathan Siemers|Associate Director|Applied Genomics|Bristol-Myers Squibb Pharmaceutical Research Institute|HW3-0.07|P.O. Box 5400|Princeton, NJ 08543-5400|(609)818-6568|nathan.siemers@bms.com