slow data transfer over nfs ... is it fedora?

[Douglas Furlong]

On Thu, 2003-12-04 at 20:02, Harry Putnam wrote:
> I lack much experience with nfs but I suspect I should be seeing much
> better transfer rates here.  Wondering if a few posters could post a
> timed transfer of know amount of data
> 
> This is from a rh9 to yarrow (fully updated)  I'll show exports file
> at the end:
> 
>    root # du -sh $rea/News/agent/nntp/enews*
>   169M    /home/reader/News/agent/nntp/enews.newsguy.com
> 
>    root # time cp -a  $rea/News/agent/nntp/enews*/ /EXP_root
>   real    12m26.120s
>   user    0m0.340s
>   sys     0m13.990s
> 
> So 169MB in 12 1/2 minutes.

This is inordinately slow speed, have you tried scp/ftp/smb to transfer
the same file? This will help narrow the problem down from an NFS issue
to a network issue.

What does tail -f /var/log/messages show? I had link errors in the past
due to what I can only assume was a failing network card (had neither
the skill not patience to figure out if it was or not).

To give you an idea of transfer rates at work.

time dd if=/dev/zero of=/home/douglas.furlong/test bs=16k count=16384

The above results in a 256MB file being transfered, I get better results
at home, as I am suffering NFS performance issues at work which I am in
the middle of eliminating.

real    0m54.698s
user    0m0.030s
sys     0m0.980s

real    1m5.801s
user    0m0.010s
sys     0m0.880s

real    1m0.551s
user    0m0.050s
sys     0m1.000s

The quickest there is 54 seconds to transfer the 256MB file.

> cat /etc/exports:
>   /                        reader(rw,async,insecure_locks,no_root_squash)
>   /home                    reader(rw,async,insecure_locks,no_root_squash)
>   /var                     reader(rw,async,insecure_locks,no_root_squash)
>   /mnt/exp                 reader(rw,async,insecure_locks,no_root_squash)
>   /usr                     reader(rw,async,insecure_locks,no_root_squash)
>   /tmp                     reader(rw,async,insecure_locks,no_root_squash)

A more useful piece of information would be the mount commands that you
use.
Placing rsize=8192,wsize=8192 in the options part of the mount provides
better perfomance than the default values, however I am guessing they
are already there as they are listed in the man pages for the mounts.

neon:/mnt/raid/home     /home                   nfs    
retrans=10,wsize=8192,rsize=8192,intr,hard 0 0

Try using the nfsstat command, you can get some information there with
regards to errors and "retransmits" saying how many times the packets
need to be resent.

I hope this gives you some help.

Doug