Re: MVCC for massively parallel inserts

From: Greg Stark <gsstark(at)mit(dot)edu>
To: "Joshua D(dot) Drake" <jd(at)commandprompt(dot)com>
Cc: Greg Stark <gsstark(at)mit(dot)edu>, pgsql-general(at)postgresql(dot)org
Subject: Re: MVCC for massively parallel inserts
Date: 2004-01-06 06:15:53
Message-ID: 87k7451u1y.fsf@stark.dyndns.tv
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

"Joshua D. Drake" <jd(at)commandprompt(dot)com> writes:

> Something I have been toying with is getting two of the 12 drive 3Ware cards
> and running RAID 0+1 across them (with LVM). At just under 300 for the cards
> and only 80 bucks a drive (80 Gig)... that is alot of space, and a lot of
> speed for not a lot of money.

Unless I have things backwards, raid 0+1 means if any drive fails the whole
side of the mirror fails. if two drives fail you have a 50/50 chance of the
them being on opposite sides of the mirror and losing the whole thing.

Even if you don't have a double failure, resyncing seems like it would be a
pain in this situation. LVM wouldn't know about the stripe set so it would
mean resyncing the entire 12-disk array on the failed side of the mirror.

I thought it was generally preferable to do Raid 1+0 (aka "raid 10") where any
two drives can fail and as long as they aren't precisely opposite each other
you're still ok. And resyncing just means resyncing the one replacement drive,
not the whole array.

--
greg

In response to

Browse pgsql-general by date

  From Date Subject
Next Message Randall Smith 2004-01-06 07:10:28 cross-table unique constraint
Previous Message Bruce Momjian 2004-01-06 05:28:39 Re: release notes/Appendix E in documentation