Re: RAID on Hotspots....

rich wrote:

> All,
>
> It seams that in the Oracle on Linux shop i/o doesn't follow
> tradition.

That's because tradition was dealing with the technology of the time, which probably no longer applies for most.

> It used to be that you always separated things physicaly
> on different controllers, different disk, etc... Today it's just one
> big RAID array and everything goes on it. How many of you have 10 EMC
> arrays? I'll bet not many, if any. How are you separating data from
> index,

Hopefully not many, since there's no point, and doing so addresses the wrong issue. Indexes and their tables don't intrinsically contend for I/O any more than two tables or two indexes do. The real issue is therefore to find out what *actually* contends, not what hoary old myths say should contend.

> redo logs from arhived redo logs,

I agree with this one, though the real issue is that the online redo logs should be separate from *everything* else, not just archives.

But take a look at 10g and ASM. The whole point is that very few shops can afford the time or the costs associated with hiring truly skilled DBAs to fine-tune this sort of stuff on an on-going basis. So 10g is pushing into the 'give me enough spindles and I'll handle it so it might not be perfect, but it will be close enough you won't notice' database practicioner school of thought.

For most people most of the time, that approach is fine.

Do a search for "Stripe and Mirror Everything" on Google.

And since you specifically mentioned Linux, take a look at what's coming in Fedora Core 3. Their Test 2 release dispenses with boring old things like creating "/", "/usr", and swap partitions, and wraps the entire hard disk inside a single logical volume manager. It's ASM for the non-Oracle crowd, and took me by surprise when I saw it. But it's happening, and why not? I bet a smart O/S can sort these sorts of things out better than I ever could.

Regards
HJR
> temp from undo?
>
> Please send your solutions to rheadrick_at_tickleinc.com
>
> Thanks,
>
> Rich
Received on Tue Oct 05 2004 - 17:14:02 CDT