On 5/1/2011 12:10 PM, prad wrote:
neither is this an argumentum ad antiquitatem. again all that is being shown is that xfs has a long history of use again with a reputable organization. again, it is merely supporting evidence and it is not being argued that because the organization has used this for x years, it should necessary continuous to do so because of this fact.
And in fact, for those who did not read further on the NAS website, Columbia is no longer the prime compute resource at the NASA Advanced Supercomputing Division at NASA Ames Research Center. C/XFS are no longer the primary filesystems used by the main compute resource. CXFS simply can't scale to the client node count and bandwidth of the Pleiades cluster--it wasn't designed to. Columbia had 20 fat nodes of 512 CPUs each, 20 CXFS clients. Pleiades has 10,752 thin nodes. The only suitable production filesystem for such a high node count compute cluster machine is the Lustre cluster filesystem, specifically designed for large node count clusters.
-- Stan