I believe that following items are more important than solving
performance problems with buying expensive hardware:
* data-model (good normalization)
* tuning queries (forcing using the correct indices if the engine isn't
* Use stored procs
* balancing page-buffers, sort-memory etc vs total memory on server
* separate data, journaling files (IB2007) and temporary files (sorting,
etc) onto different disk-controllers
* If reports are a problem, try to create replication DB(s) and redirect
your reports onto these DB. So you can tune the original DB for writing
(OLAP) and the other DB for data-mining.
* If you have too much concurrent user, write a n-tier application with
connection-pooling instead of fat-clients
I have a lot of cases when customers have performance problems, and
after revising some structures (indices, create stored procs, etc..) and
queries the problem is solved. It is rare that you need a very powerful
server, if you need that then it is probably time too pay for Oracle, DB/2.
Raid 5 isn't very good at writing, so if your application does a lot of
DML-statements, then it is better to use RAID 10. Best of 2 worlds,
fast reading/writing and mirroring.
Van den Wouwer Danny