RE: Miserable Disks
- From: "Matthew Zito" <mzito@xxxxxxxxxxx>
- To: <henry@xxxxxxxxxxxxxxx>, <jkstill@xxxxxxxxx>
- Date: Sun, 4 Jun 2006 01:45:56 -0400
The way the Clariion handles its write cache is that a certain amount of the
first set of disks (maybe a a gig each of the first 6-ish drives? I don't
remember) are reserved as destage areas for the write cache. There's a battery
backup system for just those drives (more practically, the first disk
enclosure) and the SPs, such that you don't need to have an enormous battery
system for potentially hundreds of disks (a la the symmetrix). The write cache
is internally mirrored to handle RAM failures. Upon power failure, the
Clariion SP will destage the pending write cache to the stage area on the disks
that are being battery backed (in a few different places). Once the power is
restored to the array, it will move the outstanding writes from the stage areas
to the real data areas on disk.
Now, the real question is: is there a secondary battery to keep the RAM powered
in the event of a catastrophic power outage? My guess is no, but I don't
remember for sure.
Regardless, a graceful powerdown of a clariion should result in an orderly
cache destage as part of the powerdown cycle. Even one done a little
incorrectly shouldn't cause a loss of data. It might be safer to disabe the
write cache pre-powerdown, but it sounds like an edge case or a bug you hit.
The problem with all cache systems is that they're complicated. Cache can hide
a multitude of sins, and consequently there's a lot of logic built in to
heavily optimize, making them likely sources of bugs.
Thanks,
Matt
--
Matthew Zito
Chief Scientist
GridApp Systems
P: 646-452-4090
mzito@xxxxxxxxxxx
http://www.gridapp.com
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx on behalf of Henry Poras
Sent: Sun 6/4/2006 1:17 AM
To: jkstill@xxxxxxxxx
Cc: pythianbrinsmead@xxxxxxxxx; oracle-l@xxxxxxxxxxxxx
Subject: Re: Miserable Disks
I asked our SA about this and he said that the write cache is non
volatile as long as it is enabled and the backup battery system has
power. This means that the write cache should be small enough to allow
it to flush its contents before the battery depowers, but given this,
there is nothing to worry about.
As I am still learning EMC stuff, is this true? Are there other things
that can go wrong with write cache?
Thanks.
Henry
Jared Still wrote:
> Comments inline.
>
> On 5/25/06, *Mark Brinsmead* <pythianbrinsmead@xxxxxxxxx
> <mailto:pythianbrinsmead@xxxxxxxxx>> wrote:
>
> (*) Get the biggest cache you can afford. (Make sure it's
> *really* non-volatile. Wolfgang
> never mentioned what happens to your database when you lose the
> contents of the write
> cache, but *I* got a taste! And it wasn't pleasant!)
>
>
> Yes, I had experience with that 2 weeks ago on a CX700.
> Definitely unpleasant.
>
> EMC claimed that we should have disable the write cache
> prior to powering down the SAN.
>
> I have since learned that the support tech in question was blowing smoke.
>
>
> --
> Jared Still
> Certifiable Oracle DBA and Part Time Perl Evangelist
--
//www.freelists.org/webpage/oracle-l
Other related posts:
