chris dagdigian wrote: [snip] > (2) Diskless. I would never _not_ put a large cheap IDE drive in my > cluster elements because at the very least I need a local large /tmp or > /scratch partition for caching sequence databases and raw data. Without > this approach you pretty much will swamp your NFS server under any sort > of serious workload. [snip] I agree completely. If the reference sequence data is split and stored at each node, then only the query sequences and results (or parsed results) need to move in and out of the NFS server and across the network (well, each ref db update moves once). No need to swamp the NFS or cause massive collisions on your network. I am interested, has ANYONE used a diskless BLAST cluster? -- Eric Engelhard - www.cvbig.org - www.sagresdiscovery.com