Re: Naive view of Oracle on RAID Subsystem?

The point mentioned earlier in this thread regarding total RAID system redundancy is well made. Yes, you may be guarded against a single disk failure, but if you have only one controller to access the RAID disk ranks, and that controller fails, you're done.

My group has an NCR 3555 running UNIX SVR4, with an NCR 6298 RAID cabinet. We're currently using ORACLE v6.037, and in the process of migrating to v7.1.3. Our RAID cabinet has two ranks, each with 5 2Gb disks. The optimal distribution of things that we've found looks like this:

• ORACLE executables on one internal disk
• Redo Logs on another internal disk
• Rollback tablespace on another internal disk
• Data tablespaces on one RAID rank
• Index tablespaces on the other RAID rank

       Actually, the layout above of the RAID ranks is oversimplified. We've tried to balance the data and index tablespaces between the two RAID ranks, based on our users queries.

        I am also going to move our default tablespace, TEMP, out of the RAID ranks, onto a separate disk.

        BTW, everything not in RAID is being mirrored to another disk with a SVR4 tool called Volume Manager. We use raw slices for everything, redo logs and tablespaces.

Lou Avrami ( attmail!lavrami )

Opinions stated are my own, not my company's. Received on Fri Dec 30 1994 - 17:55:21 CET