[PLUG] NFS write performance?

Paul Heinlein heinlein at madboa.com
Thu Dec 4 08:08:02 UTC 2003


On Thu, 4 Dec 2003, Russell Senior wrote:

> Russell> Hmm.  I am evidently doing something wrong.  I get an order
> Russell> of magnitude better write performance from an scp compared to
> Russell> a write over an nfs mount.  I am just looking at "blocks out"
> Russell> from a "vmstat 1" running on the server.  I see, roughly,
> Russell> 300-400 with NFS and and about 4000 from an scp.  I am not
> Russell> sure if those are 1k blocks or 4k or something else, but the
> Russell> order of magnitude stands out pretty clearly.
>
> Chris> Well, what options are you using to mount? And what options in
> Chris> /etc/exports?
>
> mount options are: rw,sync,hard,intr,addr=192.168.0.55
> /etc/exports options are: rw,sync
>
> The server is running a 2.6.0-test11 kernel, the clients are mixed
> 2.4.2x and 2.6.0-test11.

Can you repeat the test with the server booted into a recent 2.4
kernel? On the nfs mailing list, there have been repeated attempts to
figure out some 2.6-specific write problems. Trond's current hunch is
that the 2.6 VM mechanism has a bug that's getting tickled, but no one
knows for sure yet.

--Paul Heinlein <heinlein at madboa.com>




More information about the PLUG mailing list