>From looking at this quickly it sounds exactly like mbuf exhaustion. I believe 
these are fixed in number (they were under the RiscOS port of the BSD 
networking code :) [1]

I dont recall how, when I was looking at this area of NetBSD, I changed it 
(there is a config option or rather a way to get config to do it) - but I know 
I ran a machine for a while that was just a network client with 64 mbufs. 
Sounds like you need to find the default current amount and double it (or if 
you're short on memory add a few - remember they are statically allocated!).

A good place to start looking would be sys/kern/uipc_mbuf.c

Paul. 

[1] Remember the RiscOS !Internet is just a port of the BSD4.3/4.4 network 
layer - not that I would know :)


S.J. Borrill wrote on Wed, 15 Jul 1998 07:33:54 GMT:

     > I've a RiscPC with two ethernet cards in running 1.3F. The machine is
     > acting as a web proxy. There are a number of client machines on one
     > interface and the other interface is conneted to an ISN router. When I
     > have more than a couple of pages being fetched the connection gets slower
     > and eventually hangs. If I try to ping anywhere , I get the error 'No
     > buffer space available'. netstat shows thaht about 38 sockets are being
     > used and much of these are in a wait state. After a few minutes, the tcp
     > connections time out, the sockets are freed up and everything works again
     > for a small while. I presume this is like mbuf exhaustion under RISC OS.
     > How do I get around the problem?
     >
     > Stephen Borrill
     >
--
Paul Wain
Development Manager, NCOS (Corporate Client/Server)
Network Computer Inc, CA, USA

This message was sent from a StrongARM based n|c