Ramon van Alteren wrote:
Michiel de Bruijne wrote:
Where you able to resolve this?
Yep I have been able to resolve this by switching to the userspace
nfsd implementation
It appears to be a kernel bug with regard to the combination of LVM2 +
NFS-kernelspace + 9Tb
It turned out to be another bug related to reiserfs.
For those interested more data is here [1]:
http://sourceforge.net/mailarchive/forum.php?
forum_id=4930&max_rows=25&style=flat&viewmonth=200511&viewday=1
and here [2]:
http://marc.theaimsgroup.com/?l=reiserfs&m=112255940804474&w=2
In short: Although reiserfs reports in it's FAQ that it supports
filesystems up to 16Tb with the default options (4K blocks) it supports
only 8Tb. It doesn't fail however and appears to work correctly when
using the local filesystem, the problems start showing up when using NFS.
From the post on [2] this is caused by a bug in reiserfsprogs that
doesn't fully check the blockdescriptor when formatting. Apart from that
the FAQ entry stating that reiserfs supports filesystems up to 16Tb is
just plain wrong.
I fixed the problem by using another filesystem. Based on comments on
the nfs-ml and the excellent article in last months linuxgazette on
filesystem performance we switched to jfs.
So far it's holding up very good. We haven't seen the problems
reappearing with files in excess of 1Gb
Will be stress-testing nfs probably this week. If anybody is interested
I could post the data ?
Regards,
Ramon
--
To be stupid and selfish and to have good health are the three requirements for
happiness, though if stupidity is lacking, the others are useless.
Gustave Flaubert
--
[email protected] mailing list
