Hi Thomas,

On Feb 10, 2010, at 16:51 , Connie Sieh wrote:

> On Wed, 10 Feb 2010, Matthias Schroeder wrote:
> 
>> Thomas Kress wrote:
>>> Hello All,
>>> 
>>> at RWTH Aachen, Germany, we are experiencing frequent kernel
>>> freezings of our (a bit older) data servers when using (SL5,) the xfs
>>> file system and NFS4 with i386 data server architectures. No problems
>>> with 64 bit servers.
>> 
>> You are aware that xfs is not really supported on 32 bit systems?
> 
> The stack issue is the biggest reason that we state that 32 bit systems are not supported.  The other reason is that any SGI patches are only for x86_64 .

The x86_64 module is now also coming with TUVs kernel, and is much more recent. Our (or CentOS') external kernel modules have worked very well, but personally I'm more comfortable using the ones from TUV.

>> 
>>> 
>>> From the /var/log messages we think that the problem is due to a
>>> smaller stack size for the i386 kernel as described here:
>>> http://www.mythtv.org/pipermail/mythtv-users/2005-May/089314.html
>>> 
>>> Any chance that for future SL kernel upgrades 8k instead of 4k stack
>>> size is used also for the 32 bit architecture (kernel parameter
>>> CONFIG_4KSTACKS=n)
> 
> This change was done by TUV and we are not going to change it unless they do.  And I suspect that to not happen.
> 
>> 
>> Epsilon, I would say.
>> 
>>> or any idea how to cure the problem ?
>> 
>> Go to 64 bit.
>> 
> 
> XFS works better on 64 bit
> 
>>> In our
>>> opinion ext3 is not a very good choice for big data servers in case
>>> of fs crashes.
>> 
>> Why insist on a 32 bit system for 'a big data server'???
>> 
>> Matthias

and why is xfs so much better than ext3 if your fs crashes? Neither of them requires an fsck in that case. I'd much rather go with ext3 than run xfs on i686.

Cheers,
	Stephan

>>> 
>>> I found this issue discussed already in Jan 2006 on this list but I
>>> wonder whether there was any progress during the last four years.
>>> 
>>> Thanks & cheers, Thomas.
>>> 
>>> -- Mit besten Gruessen/With kind regards,
>>> 
>>> Thomas Kress, RWTH Aachen, III. Physikalisches Institut, Lehrstuhl B
>>> Office Aachen:  28A 206, Phone: +49 241 80 27281, Fax: ... 22244
>>> Office CERN: B40 4-A16, Phone: +41 22 76 71682,   Fax: ... 78940
>>> Email:
>>> [log in to unmask]<mailto:[log in to unmask]>
>>> ; [log in to unmask]<mailto:[log in to unmask]> Signed with a
>>> certificate issued by GridKa-CA (GermanGrid)
>>> 
>>> 
>>> 
>>> 
>> 
> 
> -connie sieh


-- 
Stephan Wiesand
DESY -DV-
Platanenenallee 6
15738 Zeuthen, Germany