Re: e1000 intel driver bug (which impacts nfs)

[Trond Myklebust <trond.myklebust@fys.uio.no>]

>>>>> " " =3D=3D Thomas Lang=E5s <tlan@stud.ntnu.no> writes:

     > We've also got problems with nfs-hangs when transfering large
     > files (ie.  files around 300-400M, sometime we have to go a bit
     > higher tho, like 2GB-3GB files, but it's always possible to
     > trigger this. However, we don't need to be jumping through
     > hoops to "fix it", after a min or so, it's ok again. It seems
     > to me like there's a VM problem or something.

     > We've got 2GB mem on the machines which are suffering from
     > theese problems.

There is a problem with highmem that I'm just barely getting to grips
with: NFS is able to starve the kernel of highmem bounce buffer
resources because we kmap() the pages for too long (surprise: NFS
predates the highmem code by several years and so nobody ever
considered kmap() when the design was made).

An attempt at a fix has been merged in to the development kernels as
of 2.5.19. The same patches are available for 2.4.19-pre8 +
NFS_ALL. If you'd like to test it out, you will need

http://www.fys.uio.no/~trondmy/src/2.4.19-pre8/linux-2.4.19-NFS_ALL.dif

+ the 4 patches from

http://www.fys.uio.no/~trondmy/src/2.4.19-pre8/alpha

Cheers,
   Trond

_______________________________________________________________

Don't miss the 2002 Sprint PCS Application Developer's Conference
August 25-28 in Las Vegas -- http://devcon.sprintpcs.com/adp/index.cfm

_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs