poor NFS Performance - improves when remounting

Wolfgang Schludi <(E-Mail Removed)> wrote:

> The NFS write performance after booting is very bad (~300 kBytes/sec).
> The network (100Mbit/sec; full duplex) seems to be ok since both "rcp"
> and "ftp" run at ~10.9 MBytes/sec in both directions.

> excerpt from /etc/fstab@client:
>
> fileserver:/filetransfer /filetransfer nfs bg,hard,intr,nolock,noac 0 0

I think perhaps you missed a _BIG_ optimisation you can make. Add the
rsize=8192,wsize=8192 to your mount options.

Why this isn't the default? I can't correctly recall, but you might find
the answer at nsf.sf.net

--
Cameron Kerr
(E-Mail Removed) : http://nzgeeks.org/cameron/
Empowered by Perl!

Hi,

we have a SuSE 9.0 NFS server (SuSE 2.4.21-215-default kernel;
nfs-utils-1.0.6-6) and a SuSE 9.0 NFS client (SuSE 2.4.21-215-smp4G
kernel; nfs-utils-1.0.6-6).

The NFS write performance after booting is very bad (~300 kBytes/sec).
The network (100Mbit/sec; full duplex) seems to be ok since both "rcp"
and "ftp" run at ~10.9 MBytes/sec in both directions.

After a "mount -oremount /nfs-dir" the write performance increases up to
5.5 MBytes/sec (which is acceptable for our environment).

I read the "Linux NFS-HOWTO" and tried most tips from "Optimizing NFS
Performance", but there were only minor performance changes.

------------------------------------------------------------------------

questions:

- why is the initial performance so bad?

- what does a remount change?

- is it normal (at least under linux), that the NFS write performance is
about half the speed as "rcp" or "ftp"?

- did anybody else notice this "feature"?

- any cure?

------------------------------------------------------------------------

info:

excerpt from /etc/exports@server:

/filetransfer
213.209.132.192/255.255.255.192(rw,no_root_squash,no_subtree_check ,sync)

excerpt from /etc/fstab@client:

fileserver:/filetransfer /filetransfer nfs bg,hard,intr,nolock,noac 0 0

------------------------------------------------------------------------


regards,
schludi

Cameron Kerr wrote:

> I think perhaps you missed a _BIG_ optimisation you can make. Add the
> rsize=8192,wsize=8192 to your mount options.

not really. I tried these options and noticed only a very minor
performance improvement (~2%). ok, I didn't do real performance analysis
(i.e. bonnie, IOzone, ...) but just copied a 1GB file to the NFS device.

the main question is:

- is it normal, that NFS writing takes about twice as long as e.g. ftp?

and finally: the really BBIIGG optimization came from remounting (factor
~20). And that seems to be a bug.

> Why this isn't the default? I can't correctly recall, but you might find
> the answer at nsf.sf.net

Well, I have no problem with changing the defaults in fstab/exports, as
long as they really give me a performance boost.

regards,
Schludi