NFS bitches...

Robert G. Brown
Wed Jul 29 10:15:55 1998

On Tue, 28 Jul 1998, Ravi Swamy wrote:

> Even with 8k rsize,wsize I only get around 300kB/s.  This is on a PPro 200 w/
> a 3c905 on a 100Mb hub.  There is very little traffic on the hub, I can
> run the test at 1am and see very little difference.

Ditto, and with a variety of networking cards and even using a linux
server -- it's even worse using a Sun server.  We have switched 100BT
connections, not a hub, so network bandwidth is completely out of the
picture as a rate limiting parameter.

> > IMO, the slowness problem is in the linux NFS code, because solaris,
> > irix and osf/1 (and possibly others) can read/write to/from each other
> > quickly.
> I've never seen it explained.  I always hear people say "change the rsize/wsize"but it is just a small improvement.  I can ftp files to and from Linux<->Solaris
> and get almost 8MB/s but NFS writes crawl...

This is a serious problem.  I've brought it up time and again on
various linux lists.  The usual response is "change the rsize and
wsize" but obviously YMMV far more than is reasonable and nobody seems
to know why.  Linux NFS is "broken" in other ways, as well.  I've
tried making big packages (perl, actually) on a Sun mounting the
sources from a linux server, only to have the make "break and fail"
because the caching was broken on the linux side.  The Sun was unable
to obtain correct stat information on various files and links in a
clearly time dependent way (that is, one time you could run ls and get
a correct stat, another it would fail utterly!).  Obviously a serious
problem to anybody contemplating using a Linux NFS server in a
partly-anything-else environment!

For better or for worse, Linux is a user-supported environment.  I
don't have time to figure out NFS and fix the code, and therefore
don't have the right to truly bitch about this (I reserve the right to
point out the problem repeatedly so that those who claim that there
"is no problem" don't get complacent).  I do, however, really truly
wish that NFS would be properly fixed in linux so that hand tuning
rsize and wsize is unnecessary, so that linux/solaris NFS writes and
reads are no worse than solaris/solaris writes and reads in comparable
circumstances, and so that linux can server solaris reliably, WITHOUT
your milage varying from installation to installation or even system
to system.


Robert G. Brown	             
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525