I'm wondering if I am the only one that runs into this problem:

I have a diskless sparc (SS2) used as a router. It ran stable forever using
a 1.6 kernel. Since upgrading to a newer version of the 1.6 branch it seems
to have severe problems related to kernel memory usage (see PR 20328).

This is probably not related to the network traffic (or amount thereof), as
(1) all of the traffic it sees go to a second router (same source,
i386, and at the NFS server for the sparc) too. This i386 machine has not
had *any* problems. (2) routing still continues to work, when the problem
starts to appear, for a while. But new processes can not be forked (ssh's,
or cron jobs - there is nothing going on on that machine realy). There are
no log entries whatsover - I mostly notice it when I can't log in via
ssh or the serial console. (syslog goes to the nfs server too)

This makes me suspect either some sparc specific problem (unlikely, there
haven't been any significant changes vs. 1.6) or a problem of the NFS client
code - or a network driver specific problem and some strange combination
of events that I don't fully understand.

The network interfaces involved are: gre0 (low traffic tunnel), pppoe0
(main connection to the internet) using le1 as the ethernet connection
to the DSL modem, le0 (the internal interface) on that machine is dead,
so not used, and be0 @ qec0 as connection to the other router and NFS
server.

I'm about to downgrade this machine to 1.6 and see if that makes it stable
again (or if I just have been lucky before).

Any hints apreciated.

Martin