On Sun, 2012-11-11 at 07:18 -0500, Karsten Suehring wrote:
> Package: src:linux
> Version: 3.2.32-1
> Severity: important
>
> I originally niticed the issue on Ubuntu 12.04.1, but could reproduce it on
> Debian as well.
>
> I upgraded a squeeze system to testing and found that NFS access
> creates a high load on the NFS server. A single client machine with
> a single write causes 40% system load, where the original load on Squeeze
> was 7-10% (which still is high, but might be causes by running in a test VM).
I don't think load figures from a VM are likely to be meaningful.
[...]
> I have played around with different nfs export and mount options. The only
> change that I found was using -o proto=uds when mounting. This reduces the
> load to 15% which is still higher than the original load, but still much lower
> that 40%.
>
> Ubuntu Bug #879334 and it's duplicates report the same issues.
This says the problem was bisected to:
commit 9660439861aa8dbd5e2b8087f33e20760c2c9afc
Author: Olga Kornievskaia <aglo@citi.umich.edu>
Date: Tue Oct 21 14:13:47 2008 -0400
svcrpc: take advantage of tcp autotuning
so no wonder UDP doesn't suffer as much. (But I can't recommend ever
using UDP for NFS.)
It also refers to upstream bug
<https://bugzilla.kernel.org/show_bug.cgi?id=40912> where the following
fix is suggested:
> BTW, could you test the latest upstream? This may be fixed by
> d10f27a750312ed5638c876e4bd6aa83664cccd8 "svcrpc: fix svc_xprt_enqueue/svc_recv
> busy-looping".
[...]
That fix was backported into 3.2.29, so we have it already. Are you
definitely running the above version (3.2.32-1) on the server?
Ben.
--
Ben Hutchings
Design a system any fool can use, and only a fool will want to use it.
Attachment:
signature.asc
Description: This is a digitally signed message part