🚀 go-pugleaf

RetroBBS NetNews Server

Inspired by RockSolid Light RIP Retro Guy

Article View: comp.arch.storage
Article #445

Re: RAID 1 or 5 ?

#445
From: Kevin Strietzel
Date: Tue, 03 Aug 1999 00:00
56 lines
3088 bytes
highend wrote:
> In article
> <EF4B31908F2A7BAC.E7C81737A020640C.DF7A420979E4C82E@lp.airnews.net>,
> malc@cwix.com says...
> >On Thu, 29 Jul 1999 12:01:25 +0200, Ren� K�hnen Wiesemes
> ><koehnen.pad@sni.de> caused to appear as if it was written:
> >
> >>none of your suggested solutions will work
> >>for 2 simultaneously HD failures:
> >
> >True.
> >
> >>RAID 1 (stripping) gives none fault tolerance at all.
> >>If you mirrored two RAID 1 groups, only one part
> >>can be out of order without data loss.
> >
> >You're confusing RAID0, which is striping, with RAID1, which is mirroring.
> >
> >With the suggested scheme, drive 0 would be mirrored with drive 2, drive 1
> >with drive 3, then stripe the two pairs together.  You can lose the *right*
> >two drives (e.g. 0 and 1 or 3, 1 and 0 or 2, etc). but not the wrong pair
> >(i.e. 0 and 2, 1 and 3)
> >
> >>> If I am interested in having storage fault tolerant up to
> >>> 2 x HD failure, which option will be good ?
> >>>
> >>> Option #1 RAID 1 with 4 HDs, i.e. 2 pairs
> >>> Option #2 RAID 5 with 4 HDs, i.e. 3 in group, 1 standby
> 
> true
> RAID 5 would only support the loss of one drive at a time
> and i can't think of anything that would support the loss of "any"
> 2 drives at once
...

I was going to say:  Something like RAID 5, but with the parity stripes
mirrored across two drives.  But this still doesn't let you recover from
all cases.  Easiest to think about in RAID 4 terms: you can lose both
parity disks, or one parity disk and one data disk, but if you lose two
data disks you're stuck.

Then I realized:  Think about this in ECC terms.  Use enough parity
stripes/disks to implement double-bit correction.  I don't know how many
this takes, but it's relatively a lot.

Given Options #1 and #2 above, I'd probably choose #2, if I could stand
running degraded.  And as somebody else posted, the drives themselves
aren't the only things to fail.  Adapters, cabling, fans, and power
supplies also fail.

--Kevin Strietzel
--Not speaking for my employer.

(PS - Virtually all of Stratus Computer customers use only mirroring and
concatenation, but until relatively recently that's all we directly
supported.)
Technical Details
Message-ID: <37A7878F.2EF6FF63@Stratus.COM>
Path: arc1-ca.pugleaf.net!unknown!not-for-mail
References: <7nl85m$3nn$1@newton.pacific.net.sg> <37A02675.23233EB8@sni.de> <EF4B31908F2A7BAC.E7C81737A020640C.DF7A420979E4C82E@lp.airnews.net> <7nsoie$5g1$2@birch.prod.itd.earthlink.net>

Show Full Headers
Complete Article Headers
X-Google-Language: ENGLISH,ASCII
X-Google-Thread: 10e1c8,813a29bed80cb2f6
X-Google-Attributes: gid10e1c8,public
X-Google-Thread: 1144af,813a29bed80cb2f6
X-Google-Attributes: gid1144af,public
X-Google-Thread: f4d7c,813a29bed80cb2f6
X-Google-Attributes: gidf4d7c,public
From: Kevin Strietzel <Kevin_Strietzel@Stratus.COM>
Subject: Re: RAID 1 or 5 ?
Date: 1999/08/03
Message-ID: <37A7878F.2EF6FF63@Stratus.COM>
X-Deja-AN: 508580157
Content-Transfer-Encoding: 8bit
References: <7nl85m$3nn$1@newton.pacific.net.sg> <37A02675.23233EB8@sni.de> <EF4B31908F2A7BAC.E7C81737A020640C.DF7A420979E4C82E@lp.airnews.net> <7nsoie$5g1$2@birch.prod.itd.earthlink.net>
Organization: Stratus Computer, Inc.
Content-Type: text/plain; charset=iso-8859-1
Mime-Version: 1.0
Newsgroups: comp.arch.storage,comp.sys.ibm.pc.hardware.storage,fido.ger.storage