funny route table?
i recently installed woody as the base for a firewall. i basically
set up a scheme where it was the only link to a subnet of 'protected'
computers and no packets were allowed to reach them. i had set up
NAT on them and let those packets through, but the connections would be
originating from behind the wall. NFS was acting curiously slow.
to ensure that it wasn't the NAT, i opened the firewall up to forward
packets, and added a route on my NFS server to the subnet of computers
behind the (now disabled) firewall. interestingly, NFS was still
horendously (unusably) slow, although it did actually function given
enough time. i noticed that my route table on the debian firewall had
entries of '40' for the MSS, a rather odd number, considering that the
MSS is generally 40 *less* than the MTU (which is 1500 for ethernet).
i hadn't set these explicitly, and 'tracepath' reported an MTU of 1500
to UDP port 2049 (NFS). i'm no network wiz, but something was funny.
explicitly setting the MSS to 1460 (1500 - 40 for headers) changed
neither performance nor what 'tracepath' reported for the MTU.
ANYWAY, does anyone know why my route table was showing 40 as an MSS?
presumably this would have been set by some auto-detection scheme, but
the topology of the network is quite straightforward (node to hub to
node to switch to node), end-to-end. all NICs are 10 or 100 Mb, and
it's all CAT5 cabling. pretty standard stuff. aside all that, does it
seem that the route MSS is misleading and its strictly an NFS problem?
what gives?
PS: i've played around with the r/w chunk sizes on the NFS clients
(SGI), of course, accomplishing nothing.
-c
------------------------------------------------------------------------
Reply to: