Re: Followup on NFS writing

David S. Miller (dm@neteng.engr.sgi.com)
Tue, 2 Jul 1996 09:13:00 -0700


Date: Tue, 2 Jul 1996 09:50:18 -0400 (EDT)
From: Chris Chiappa <griffon+@cmu.edu>

On Mon, 1 Jul 1996, Miguel de Icaza wrote:
> I have always wondered why the default for the rsize and wsize has not
> been set to 8192 now? I remember the limits for the 1k block sizes
> were there when Linux couldn't allocate buffers bigger than some limit
> I don't remember right now (4096 bytes - headers?). This restriction
> went away with the new kmalloc code (written by Alex Bligh?).

Is the larger size universally compatibile with NFS implementations? When
I tried to "Up" the read and write sizes to an old Unix box(IBM RT PC
running AOS, a VERY vanilla BSD 4.3 port), I would get I/O errors while
trying to write to an NFS volume mounted from the RT. This could very
well be some oddity of AOS, however(It's NFS is pretty ancient...).

Hmmm, one thing I do know is that if you set rsize/wize to 8192 and
you are talking to a SunOS/Solaris box, things will be much quicker.
This is because the internal blocksize SunOS/Solaris use is 8192, and
any file operation that operates on a partial 8192 byte block will
take a performance hit.

Later,
David S. Miller
dm@sgi.com