Re: Still looking for RAID answers

On Tue, 04 Jan 2005 19:19:58 +0000, Randy Harris wrote:

> I know that there has been much discussion of RAID recently in this
> newsgroup. I hope to get a bit more advice from anyone who has experience
> to offer.
>
> Several years ago, when I first become involved in Oracle administration, I
> read much about the virtues of OFA. OFA proponents, and the Oracle DBA
> Handbook strongly stated the importance of getting the various Oracle
> objects onto different physical disk drives. More recently, at the
> suggestion of several in this newsgroup, I've read about SAME (Stripe All,
> Mirror Everything). It seems that the two are rather differing approaches,
> though both generally urge the inclusion of as many physical disk drives as
> possible. Both make it clear that RAID based on parity (RAID 3 - 5) are
> undesirable. In simple terms, I suppose, we used to say that striping was
> bad, now we say that only parity striping is bad.
>
> Because of a recent hardware failure, I am now in a position to completely
> reconfigure my StorEdge array on a Sun server, running Solaris 8. I plan to
> use 10 drives, keeping the other 2 for hot spares. If I consider OFA
> recommendations, I would keep all of the drives as RAID 1 (mirrored) pairs,
> with each pair, its own filesystem. SAME would have me use RAID 0+1,
> creating a single slice of all 10 drives.
>
> It is still not clear to me, is the SAME architecture approach (RAID 0+1)
> "better" than OFA (RAID 1), "just as good", or "almost as good" for Oracle
> performance?

If, by OFA, you mean Oracle (or Optimal) Flexible Architecture, I do not understand where you get OFA = RAID 1 - please tell me where it says that.

Anyway, I don't see a direct conflict of interest if one goes by the spirit of each, rather than the letter. Seems to me OFA is designed to satisfy several needs, including:

• protect data by disk separation
• optimize performance by I/O balancing
• simplfy management by providing a directory structure

The way I see it, SAME attempts to cover the performance optimization. My personal tests (about 3 years ago) suggest that multiple SAME groups are recommended if there is sufficient I/O controller capability. IOW, I did not find SAME to be quite as performant as hand-optimization, but in general it's not bad.

Some forms of RAID attempt to cover data protection. I strongly encourage RAID 1+0 over RAID 0+1. That way I get mirroring at the disk level, allowing individual disks to go out without losing the array. The 1+0 setup can then be used to provide SAME.

Nothing else I see attempts to provide a standard for disk and file layout. So I use OFA as the basis for my disk layout, where it counts - making sure that Control files, Redo Log files, [S]Parameter files, and the like are separated appropriately (after all, RAID does not protect against a malicious or accidental rm -rf).

As far as I can tell, one really should mix and match based on business requirements, rather than put a foot down one way or 't'other.

That said, to me SAME is a way of avoiding administration and maintenance. If you need performance, you might also need to look at *where* on each spindle things are running - better to take the fast sliver of 100 disks than take all of 10 disks.

lol/FGB Received on Tue Jan 04 2005 - 14:27:13 CST