xfs
[Top] [All Lists]

Re: Repeating fs corruption

To: ViNiL <vladimir.linek@xxxxxxxxxxxxx>
Subject: Re: Repeating fs corruption
From: Martin Steigerwald <Martin@xxxxxxxxxxxx>
Date: Wed, 21 Jun 2006 12:24:09 +0200
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <200606211151.54376.vladimir.linek@xxxxxxxxxxxxx>
References: <200606201758.20400.vladimir.linek@xxxxxxxxxxxxx> <200606202144.43329.Martin@xxxxxxxxxxxx> <200606211151.54376.vladimir.linek@xxxxxxxxxxxxx>
Sender: xfs-bounce@xxxxxxxxxxx
User-agent: KMail/1.9.1
Am Mittwoch 21 Juni 2006 11:51 schrieb ViNiL:

> > Hello ViNiL,
> >
> > do you use write caching? With kernel 2.6.16 XFS got corrupted three
> > times in one week here with writecache enabled.
>
> Of course, we do use write caching! That is, why we've bought BBU
> (battery backup unit) for every RAID card. Are you going to tell me it
> is just a waste of money?

Hello,

no. Write caching should work okay, when you use the barrier 
functionality.

From what I understand it should even work when the filesystem is not 
interrupted abnormally and the harddisk does not loose power before all 
data is written. But from my experience with three XFS crashes in one 
week with 2.6.16 and write caching enabled I do not quite yet buy into 
this cause three completely unrelated kernel crashes seem to be quite 
unlikely for me. 2.6.16 with write caching off worked stable for over one 
week. With 2.6.15 instead it seemed to just work fine with write caching 
as long as the filesystem and harddisk was shutdown in regular way.

I suggest you try 2.6.17 with has barrier functionality on by default as 
soon as you convinced its stable enough for your usage. Until then it 
might be a good idea to try without write caching and see if that helps. 
Or revert back to 2.6.15 and see if that helps, but thats probably not 
worth the trouble.

I am no expert on this and YMMV. Maybe an XFS developer can give you 
feedback that is more accurate than mine.

Regards,
-- 
Martin 'Helios' Steigerwald - http://www.Lichtvoll.de
GPG: 03B0 0D6C 0040 0710 4AFA  B82F 991B EAAC A599 84C7


<Prev in Thread] Current Thread [Next in Thread>