On Tue, 2012-01-24 at 08:01 +0100, Rik Theys wrote: > Hi Ben, > > On Mon, Jan 23, 2012 at 11:28 PM, Ben Hutchings <email@example.com> wrote: > > On Mon, Jan 23, 2012 at 10:43:18PM +0100, Rik Theys wrote: > >> Please consider enabling the new NFS ID mapper functionality in the kernel: > >> > >> CONFIG_NFS_USE_NEW_IDMAPPER > > >> Enabling this feature should prevent page allocation failure errors we frequently > >> get on our file servers, as mentioned on > >> > >> https://bugzilla.redhat.com/show_bug.cgi?id=593035 > > > The allocation failure has already been reported as #636306. I > > proposed a fix for that in <http://bugs.debian.org/636306#35>. > > I agree enabling the new feature for squeeze is not an option, but can > it be considered for wheezy/sid? I assume the new feature is the Way > Forward? Yes. > The patch in #636306 has not been tested so far? I can not apply the > patch on our production system, but I can try to see if I can > reproduce the bug on a test machine. Has the patch been sent upstream? I think I may have given it a minimal test, but until someone else gives it a more serious test I won't send it upstream. > On our production system I've also increased vm.min_free_kbytes to > 512MB, which I believed would reduce the number of times we're seeing > these allocation failures. But it doesn't seem to help all that much. > The system has 72GB ram so I could increase it to 1GB or something. > Would that make sense? I'm not sure that's going to make a difference - keeping lots of memory available does not guarantee that there will be any large physically contiguous blocks as the idmapper is demanding. Ben. -- Ben Hutchings Horngren's Observation: Among economists, the real world is often a special case.
Description: This is a digitally signed message part