lamy@sobeco.com (j.lamy) (01/18/91)
We've seen a 3240 running 4.51 with the MIPS-supplied SLIP give the message "socket: out of buffer space" some times after the other end of the link went away and came back due to a power glitch. Prior to that we had experienced several instances where an M/120 running 4.50 would behave in that manner after seeing the modems drop connection. We had originally thought that the loss of carrier might have been inducing the problem, but there was no such loss in the first case I described above. We have also seen an instance where we had no transceiver on the second interface of our Eagle board on a 6280, yet egl1 was still configured up, giving the expected error messages on the console. We ifconfiged egl1 down to keep the machine quiet and found it a few hours later rather helpless, with messages about egl0: out of mbufs on the console. We've put a transceiver back on the egl1 interface and left it up and have not (yet) seen an occurance of the problem since (let us keep our fingers crossed). So the best explanation we can come up with is that the miscelaneous daemons still have open sockets on the egl1 interface (timed, routed, etc., and eventually run the kernel out of mbufs). In either case, there would seem to be a correlation between networks going down and mbufs being lost, and the machine failing a fair bit later. Has anyone else observed anything of the kind? Jean-Francois Lamy lamy@sobeco.com uunet!sobeco!lamy Research and Development, Sobeco Technical Services, Montreal, Canada H2Z 1Y7