Re: optimal RAID configuration

I'd question a bit of your reply (and some of it I didn't quite understand, so jump in if I've misinterpreted what you said):

Gijs Wuyts wrote in message <35f05a90.3466604_at_news.skynet.be>...
>I've posted a similar request 2 months ago, and basically got this.
>Todays RAID solutions allow a good throughput, both read and write,

Er, *maybe*. Be careful. Many manufacturers use large caches to achieve the figures they quote, and many of these are not battery-backed. A power failure could completely screw up your db, so check before you buy/install. And a large db with large operations can easily negate the cache.

>albeit no as with a stand alone drive with each and every one of the
>drives coupled to there own controller, and by dividing the different
>parts of oracle (rollback, logs, datafiles, redo...) you can gain
>optimal performance.

Well, you wouldn't really have a 1 to 1 mapping between drives and controllers, that's overkill (and expensive). Striped drives will give you a good boost for the type of system you seem to have in mind - read heavy, datamart/datawarehouse style. This can be done non-RAID, or as RAID0. If you also want some degree of fault-tolerance, you'll need to mirror those stripe-sets, either by 'mirroring' hardware or software, or RAID1. A good RAID controller with let you run RAID0+1, which is both striped-mirrored drives. Again, check before you buy.

Having said all that, RAID5 (which tends to have a write penalty) may be fine. Get a controller(s) on loan and *test* a few configurations on you *own* system. That's the only way you will truly know. No-one else can answer the question for you, as every system is different.

>
>Check out Adaptecs site on there AAA133 raid solution. not the
>cheapest, but it's a good bet, I've put my money ( and my career as a
>DB1/2A on it...)

I've used the Adaptec stuff and quite like it. You might also want to look at some complete, one-box solutions (EMC, etc) if your budget will go that far.

>
>Gijs
>On Fri, 04 Sep 1998 15:51:18 -0400, Daniel Nigrin
><nigrin_d_at_a1.tch.harvard.edu> wrote:

MotoX. Received on Mon Sep 07 1998 - 02:35:48 CDT