From: | "Merlin Moncure" <mmoncure(at)gmail(dot)com> |
---|---|
To: | "Luke Lonergan" <llonergan(at)greenplum(dot)com> |
Cc: | "Michael Ben-Nes" <miki(at)epoch(dot)co(dot)il>, "PostgreSQL Performance" <pgsql-performance(at)postgresql(dot)org> |
Subject: | Re: Integrated perc 5/i |
Date: | 2007-08-16 22:14:54 |
Message-ID: | b42b73150708161514r60dc3850w97e2e48787f4761c@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-performance |
On 8/16/07, Luke Lonergan <llonergan(at)greenplum(dot)com> wrote:
>
> Hi Michael,
>
> There is a problem with some Dell "perc 5" RAID cards, specifically we've
> had this problem with the 2950 as of 6 months ago – they do not support
> RAID10. They have a setting that sounds like RAID10, but it actually
> implements spanning of mirrors. This means that you will not get more than
> one disk worth of performance whether you are performing random seeks
> (within a one disk sized area) or sequential transfers.
>
> I recommend you read the section in the Dell configuration guide very
> carefully and look for supplemental sources of technical information about
> it. We found the issue clearly explained in a Dell technical memo that I
> don't have in front of me – we were shocked to find this out.
interesting. this may also be true of the other 'rebrands' of the lsi
logic chipset. for example, the ibm 8480/exp3000 sets up the same
way, namely you do the 'spanadd' function of the firmware which layers
the raids.
fwiw, I will be testing perc 5 raid 10, 01, 00, and 05 in a dual
controller controller configuration as well as dual controller
configuration over the md1000 (which is active/active) in a few days.
This should give very good support to your claim if the arrays spanned
in software singnificantly outperform a single controller.
merlin
From | Date | Subject | |
---|---|---|---|
Next Message | Scott Marlowe | 2007-08-16 22:25:07 | Re: Integrated perc 5/i |
Previous Message | Decibel! | 2007-08-16 19:51:10 | Re: Integrated perc 5/i |