Re: RAID: Advantage or disaster?

I'm going to disagree with your assertion. If you have only a few data files, you cannot take advantage of parallel query without a striped set of some kind as you will bottleneck on a single spindle and have idle processes that are doing nothing but waiting. Now, I used to be a RAID5 is great always zealot. I've since modified that stance, but I will say this: most applications are read intensive, and it's a documented fact that RAID5 sets are very good at reading data off the disk. There is a write penalty, but with modern controllers (note: there is not a standard for implementing RAID5. It varies from manufacturer to manufacturer) and the huge caches that are available and the high speed microprocesses on the new controllers, the bottleneck is still the disk behind the array. It's just the nature of the technology. However, there are cases, and redo is one of the most prominent, where a RAID5 set is not the best solution. Depending on how often you switch log files, you may want to have the highest throughput device you can get when it comes to redo so that you aren't waiting on redo switches. Temporary space is also one of these areas as they are about 50/50 read/write. In datawarehousing applications where you may want to have large arrays, RAID5 make alot of sense (and cents <grin>).

Getting back to the original posting, I would be more concerned as to why the server is crashing. I've been using Compaq hardware with a variety of RAID sets (5, and 1+0) and I've never had a problem with either. I think he's got more of a problem with a bad driver or board than with a drive configuration, IMO. I've run a small data warehouse on a Compaq 6500 with five RAID 5 arrays, and I've NEVER had to reboot due to an Oracle or NT failure (which actually says alot considering I've got NT on the thing!)

My $0.02, and I'll give it away for free, so you got what you paid for.

Dan Peacock
DBA
Wolverine World Wide
pagesflames_at_usa.net (Dusan Bolek) wrote in message news:<1e8276d6.0112050715.6139e479_at_posting.google.com>...
> Andrew Mobbs <andrewm_at_chiark.greenend.org.uk> wrote in message news:<wXk*Yi4ap_at_news.chiark.greenend.org.uk>...
> > RAID 1+0 is wonderful, and solves most of your problems, but is expensive.
> >
> > RAID 0 reduces the disk reliability but is great for performance.
> > Don't use it unless you're willing to assume the risk.
> >
> > RAID 5 is OK in its place. Traditionally it's been considered to be very
> > poor for write performance but good for read, however I've recently seen
> > very impressive results with striped RAID-5 sets (RAID 5+0 ?) behind
> > a very large RAM cache (16 GB). I still wouldn't recommend it for redo
> > logs though.
>
> I do not think that especially for Oracle are a good idea any RAIDs
> with stripes (data area spreads across several discs). The best option
> for perfomance is RAID1, because you will get a good perfomance and
> you always know when each part of your data is, that's something very
> hard to achieve with stripes, especially with RAID5. So using one
> RAID5 set for all database files is a very bad thing for I/O
> perfomance.
> I usually advise in those situations: plug as many disc as possible,
> mirror them and then spread all your files properly across them and
> your Oracle instance will be happy.
>
> --
> _________________________________________
>
> Dusan Bolek, Ing.
> Oracle team leader
>
> Note: pagesflames_at_usa.net has been cancelled due to changes (maybe we
> can call it an overture to bankruptcy) on that server. I'm still using
> this email to prevent SPAM. Maybe one day I will change it and have a
> proper mail even for news, but right now I can be reached by this
> email.
Received on Wed Dec 05 2001 - 16:00:03 CST