I am soliciting opinions on something. I have a small server maintaining about 300 filesystems in 30TB on one client running Linux. I have a very fast (dual Xeon with 4GB memory) Linux server running with a very fast Linux client (single 3GHz class machine) all on a private gigE network. Once things stabilize, I figure about 250-500GB per evening will be backed up, eventually to LTO1 by way of the disk storage pool(s)..
I have the following storage pools: DISK - 250GB local 15K RPM disk SLOW - 1TB+ NFS-mounted over gigE TAPE - LTO1 drives Currently, I'm testing backing up to the fast disk and having the slow disk as the next storage pool and then finally to the tape. The NFS mounted disk runs at about 35-50% of the speed of the local disk (when comparing backup time and migration time). The SLOW pool is basically free surplus equipment. The local fast DISK pool is expensive and scarce. The TAPE drives are just generic manual LTO drives. The SLOW pool is being leveraged for its cache value (pun intended) and I'm trying to use it to have the maximum number of recent file backups present online (as opposed to only 250GB online using the fast disk). Assuming that there are no bottlenecks (backups feed the fast DISK storage pool and it's able to feed the SLOW pool before it fills up), is this a sane architecture? As Wanda and others mention, the server and client are spending a great deal of time doing database transacations so I don't think the data stream is going to overwhelm any of these storage pools.
