Derek has posted his experience...I am forwarding to the list in case
some else runs into similar frustration. Although as a non-linux person
what he is suggesting went over my head I am sure his logic and what I
experienced are correct.
--
Sudev Barar <[EMAIL PROTECTED]>
--- Begin Message ---
I posted a similar problem a few weeks ago. WHat I think is going on is
that in a switch where you mix modes (100 and 10 MB) the linux NFS
causes a problem. I have done some tests and it appears that if I use a
switch with congestion control (it forces collisions with upstream UDP
packets) if its buffer used to convert 100 MB to !0 MB are full, then
things work OK. Most cheaper switches do not support this.
I puzzled over this a lot and I think the problem is in the Linux NFS
stack. I had a similar problem with an IBM Networkstation that I was
booting from NFS and it wouldn't work unless I reduced the NFS block
size down to 1024 ( < 1540 ethernet packet). Seems like UDP
fragmentation is not handled in these switches very well. I have a good
surplus IBM switch (8271-model 524) which works well, but a Netgear that
doesn't. If you have a lot of 10Mb stations you may want to look at
some of these switches and group all your low speed stuff on them.
(pcsurplusonline.com has a few brand now ones for $50).
I never got a response on the list so I presumed that it wasn't a common
problem.
Hope this helps.
Derek McKay
--- End Message ---
