Alfio Puglisi wrote:
former will be cheaper. But it seems that our disk subsystem is hampered by latency - lots of small reads looking for DB indexes, articles, things that got pushed out from the cache, all of them from multiple concurrent
If things get pushed out of the cache, you need RAM, not SCSI disks. And you don't need 40 GB of RAM, you only need so much that things don't get pushed out of the cache so often. Today's ATA disks are far better than any SCSI disk of a few years ago. And only a few years further back, people used to say that SCSI was only toys and you really needed 8, 12 or 14 inch Eagle disks, and stuff like that. A recommended read is the first chapter of "The Innovator's Dilemma".
This design-by-committee approach to hardware problems is so boring.