Re: To RAID or not to RAID ??

( please copy barry.schader_at_medtronic.com on any reply )

I'm also building a smaller (10-15GB) data warehouse using Oracle 7.3 on NT.
I'd be interested in any responses to this thread, also.

In my situation, I will have a limited number of disks (initially - five 9GB drives).
If I set them all up as one big RAID-5 set, I lose the flexibility to spread things
out manually -- to put indexes on a separate drive from the data, for example.
But RAID-5 has a lot of appeal from an availability standpoint.

Conjecture: If I opt not to do RAID at all, and spread things out manually, I
would have indexes and tables spread around in some logical fashion. If a

single query happens to be the only thing running, Oracle should be in a cycle like:
  read index blocks from drive A
  read data blocks from drive B
  repeat
I've seen Oracle doing this while executing a large query. This should result
in very minimal head movement on drives A and B, resulting in good response
time.

Now, if I were to put drives A, B, and C into a RAID-5 set, with indexes and
tables lumped together, I would see this pattern on the lone query:   read index blocks from drive A/B/C
  read data blocks from drive A/B/C
  repeat
It would seem that there could be significant head movement on all three disks, as they jump back and forth from index to data. If the data table is multiple gigabytes, this could involve some very long head movements.

It seems like the performance would be worse in the RAID-5 case. Of course,
this all goes out the window as soon as a second query starts. Then, you get a jumble of head movement anyway. But in my case, with a user population that is unlikely to reach more than 100 people, there could be a lot of queries that run alone on the machine.

I don't have the flexibility at this point to do any kind of benchmark. Anybody
have any info on this???

Thanks,
Barry Schader
barry.schader_at_medtronic.com Received on Wed Apr 23 1997 - 00:00:00 CDT