Re: RAID Controllers

From: Greg Smith <greg(at)2ndQuadrant(dot)com>
To: pgsql-performance(at)postgresql(dot)org
Subject: Re: RAID Controllers
Date: 2011-08-24 03:28:11
Message-ID: 4E546FCB.2050703@2ndQuadrant.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance

On 08/23/2011 06:42 PM, David Boreham wrote:
> I took a look at Areca. The fan on the controller board is a big
> warning signal for me (those fans are in my experience the single most
> unreliable component ever used in computers).

I have one of their really early/cheap models here, purchased in early
2007 . The fan on it just died last month. Since this is my home
system, I just got a replacement at Radio Shack and spliced the right
connector onto it; had it been a production server I would have bought a
spare fan with the system.

To put this into perspective, that system is on its 3rd power supply,
and has gone through at least 4 drive failures since installation.

> Can you say a bit more about the likely problems with the CLI ?

Let's see...this week I needed to figure out how to turn off the
individual drive caches on a LSI system, they are set at the factory to
"use disk's default" which is really strange--leaves me not even sure
what state that is. The magic incantation for that one was:

MegaCli -LDSetProp DisDskCache -LALL -aALL

There's a certainly a learning curve there.

> I'm thinking that I configure the card once, and copy the config
> to all the other boxes, so even if it's as obscure as Cisco IOS,
> how bad can it be ? Is the concern more with things like a rebuild;
> monitoring for drive failures -- that kind of constant management
> task ?

You can't just copy the configurations around. All you have are these
low-level things that fix individual settings. To get the same
configuration on multiple systems, you need to script all of the
changes, and hope that all of the systems ship with the same defaults.

What I do is dump the entire configuration and review that carefully for
each deployment. It helps to have a checklist and patience.

> How about Adaptec on Linux ? The supercapacitor and NAND
> flash idea looks like a good one, provided the firmware doesn't
> have bugs (true with any write back controller though).

I only have one server with a recent Adaptec controller, a 5405. That
seemed to be the generation of cards where Adaptec got their act
together on Linux again, they benchmarked well in reviews and the
drivers seem reasonable. It's worked fine for the small server it's
deployed in. I haven't been able to test a larger array with one of
them yet, but it sounds like you're not planning to run one of those
anyway. If I had 24 drives to connect, I'd prefer an LSI controller
just because I know those scale fine to that level; I'm not sure how
well Adaptec does there. Haven't found anyone brave enough to try that
test yet.

--
Greg Smith 2ndQuadrant US greg(at)2ndQuadrant(dot)com Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support www.2ndQuadrant.us

In response to

Browse pgsql-performance by date

  From Date Subject
Next Message Mark Kirkwood 2011-08-24 03:48:56 Re: 8.4 optimization regression?
Previous Message Tom Lane 2011-08-24 03:15:40 Re: 8.4 optimization regression?