RE: [Sheflug] NFS and fast machines

["Morris, David (Allvac, UK)" <david [at] xxxxxxxxxxxx>]

> -----Original Message-----
> From: shef-lug-admin [at] list.sheflug.org.uk 
> [mailto:shef-lug-admin [at] list.sheflug.org.uk] On Behalf Of 
> Morris, David (Allvac, UK)
> Sent: Monday, January 26, 2004 9:46 AM
> To: shef-lug [at] list.sheflug.org.uk
> Subject: RE: [Sheflug] NFS and fast machines
> 

> > If you have a spare machine, I'd be trmpted to connect it to
> > the same LAN, (by a hub not a switch), run tcpdump -n on that 
> > machine, then run tcpdump on the server, and see whether any 
> > traffic is genreated that triggers the response.
> > 
> 
> Good idea. I'll let you know what happens...

Bad form, responding to one's own emails... However...

Watching what was flying past on the tcpdump window, I spotted something
that looked strange. After a few attempts, I managed to catch the error:

"ip reassembly time exceeded"

...which appears (after much googling) to relate to packet fragmentation
with NFS. Now I'm stuffed because one suggested workaround is to reduce
the rsize and wsize parameters on the nfs mount commands... But this is
an ltsp client with a precompiled kernel and loads of other stuff... I
can sense some serious hacking here.

The other suggestion is the quality of the network cards. So far I've
used Dlink with via-rhine chipsets (DE530's) and and cheapo Netgear with
a Realtek chipset in the clients, and I believe both exhibit the
symptoms.

The server has an embedded NIC on the MB - I'm just off to check what
chipset *that* is.

-- 
D

___________________________________________________________________

Sheffield Linux User's Group -
http://www.sheflug.co.uk/mailfaq.html

  GNU the choice of a complete generation.