freebsd stable - Nov 2008 - NFS Performance Issue

Hi Listers,

We recently found that the Performance of the NFS Client in FreeBSD is
worse than that in Linux.

Linux [/net/iscsi] -jnlin- sudo ls -al
/net/iscsi/mysql/blog-2/var/pixblog_2/blogarticle.ibd
-rw-rw---- 1 3306 3306 734003200 2008-11-27 00:13
/net/iscsi/mysql/blog-2/var/pixblog_2/blogarticle.ibd

Linux [/net/iscsi] -jnlin- uname -a
Linux proxy 2.6.22.6-Linux #1 Mon Sep 3 11:54:48 CST 2007 i686 GNU/Linux

Linux [/net/iscsi] -jnlin- sudo rsync --progress
/net/iscsi/mysql/blog-2/var/pixblog_2/blogarticle.ibd /tmp/
blogarticle.ibd
   734003200 100%   15.16MB/s    0:00:46 (xfer#1, to-check=0/1)

FreeBSD [/net/iscsi] -jnlin- uname -a
FreeBSD newphp 7.1-PRERELEASE FreeBSD 7.1-PRERELEASE #3: Tue Nov 25
15:55:33 CST 2008     root@FreeBSD:/usr/obj/usr/src/sys/KERNEL  i386

FreeBSD [/net/iscsi] -jnlin-  sudo rsync --progress
/net/iscsi/mysql/blog-2/var/pixblog_2/blogarticle.ibd /tmp/
blogarticle.ibd
   734003200 100%    5.66MB/s    0:02:03 (xfer#1, to-check=0/1)

It's about 1/3 of NFS client in Linux. We have tuned TCP recv/send
buffer, and got no gain. The mount parameters are: (We use amd)
rw,nfsv3,lockd,grpid,intr,soft,cache,nodev,resvport,timeo=10,retrans=3,async,rsize=1024,wsize=1024,noatime

Does anyone has any suggestion to get the performance better? Thank you all!

> We recently found that the Performance of the NFS Client in FreeBSD is
> worse than that in Linux.
What OS is your nfs-server running?
> It's about 1/3 of NFS client in Linux. We have tuned TCP recv/send
> buffer, and got no gain. The mount parameters are: (We use amd)
>
rw,nfsv3,lockd,grpid,intr,soft,cache,nodev,resvport,timeo=10,retrans=3,async,rsize=1024,wsize=1024,noatime
>
> Does anyone has any suggestion to get the performance better? Thank you
all!
You can ommit read- and write-size using tcp-mounts. From 'man
mount_nfs':

     -w      Set the write data size to the specified value.  Ditto the com-
             ments w.r.t. the -r option, but using the ``fragments dropped due
             to timeout'' value on the server instead of the client. 
Note
             that both the -r and -w options should only be used as a last
             ditch effort at improving performance when mounting servers that
             do not support TCP mounts.

I have a solaris 9 nfs-server with vxfs and I used to mount using udp
but then I ran into 'server not responding' and changed to
tcp-nfs-mount instead. I just copied a 200 MB file to the nfs-server
in 7.5 sec. which is 26.6 MB/s.

-- 
regards
Claus

When lenity and cruelty play for a kingdom,
the gentler gamester is the soonest winner.

Shakespeare