perry@piermont.com said:
> > 	(a) Pages simply aren't being cleaned fast enough.
> >  -or-
> > 	(b) Clean pages aren't moving from the inactive list to the free
> > 	    list fast enough.
> 
> Incidently, I noticed an ugly problem with paging to local disk, too,
> last night -- when I ran out of physmem the machine went bonkers. 

I've noticed NFS swapping problems as well - there are situations
where a diskless machine simply doesn't recover from, in particular
when doing a huge compile (typilally C++) and the "scheduler - no room
for xxx" appears. Adding a local disk for swapping helped here.
Otoh, there might be general NFS problems too - it seems that a
diskless box doesn't survive an half-hour NFS server outage (v3 on UDP).
A lot of processes hang in "nfs_hashlock". (I think it was in "nfsrcvlk"
not so long ago.)
Hadn't have time to dig further - sorry.

best regards
Matthias