Terrible NFS4 performance: FreeBSD 9.1 + ZFS + AWS EC2

[Berend de Boer]

>>>>> "Rick" == Rick Macklem <rmacklem at uoguelph.ca> writes:


    Rick> After you apply the patch and boot the rebuilt kernel, the
    Rick> cpu overheads should be reduced after you increase the value
    Rick> of vfs.nfsd.tcphighwater.

Have set it to 10,000, max cpu for nfsd I've seen is below 6%. Makes
no real difference whatsoever to the great slowness of nfs4 in this
use-case.

I.e. did two tests: 17.5 minutes with sync=disabled, 21.5 minutes with
sync=enabled, but difference in this case could simply be due to
whatever else was going on that that time.

FYI, in the nfs3 mount nfsd is at 0% at all times, basically uses no
cpu whatsoever.

The weird thing is that nfs3 performance seems to have been greatly
affected: the same test which ran at 2 minutes on udp is now between 7-11
minutes.

As this could be a problem with how I'm testing now (I recompiled the
kernel), I'll try to see what numbers I get when I undo the patch and
work against a recompiled kernel.

--
All the best,

Berend de Boer


          ------------------------------------------------------
          Awesome Drupal hosting: https://www.xplainhosting.com/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: OpenPGP Digital Signature
URL: <http://lists.freebsd.org/pipermail/freebsd-fs/attachments/20130709/405ab270/attachment-0001.sig>