=> On Sat, 12 Apr 2003 10:47:26 -5, "Richard L. Rhodes" <rhodesr AT 
firstenergycorp DOT com> said:

> This is a very interesting discussion, because it's exactly opposite of what
> we do on our random access Oracle databases.  For highly random, transaction
> oriented, oracle database we strip the db as wide as possible across as many
> disks as possible.

I'm betting that the striping you're speaking of happens at a different
logical level. For example, if ORACLE knows that it has a 20GB database, with
2GB of container space on each of 10 different containers, then it is likely
that two consecutive random requests will require access from different
containers.

This is distributing the work, but only so long as each of those containers is
itself independant, perforlance-wise.

But what if all those containers are "striped" across the same two 10GB
physical spindles?

In this case you have forced all of the operations to exist in the same
performance domain, plus you have absolutely guarunteed that some 'single
block' operations will have to move multiple disk arms to be satisfied.


- Allen S. Rout