Looks like the fix is to install the latest jumbo le patch -I really should
have known.
Thanks to :
Tom Crummey, EMAIL: tom@sees.bangor.ac.uk
Glenn Satchell glenn@uniq.com.au
Marc S. Gibian email: gibian@stars1.hanscom.af.mil
alevin@ltcm.com (Avi J. Levin)
Kevin.Davidson@edinburgh.ac.uk
Problem:
>I have a SUN SS10 running 4.1.4 which suddenly stops talking to the network.
>This usually happens late at night. I can find no clues in the logs, only errors when
>things can no longer talk to NIS etc.
>
>The machine is our internal news server and I found the following message from
>nntpd just before the latest hang:
>gatekeeper.dmv.co.uk cpstdin: EOF before period on line by itself <....@.....>
>
>I can fix it by doing: ifconfig le0 down; ifconfig le0 up
A selection of replies:
------------------------------------------------------------------------------
>From Kevin Davidson
I thought this was relatively well known. The built in ethernet in
the SS10 is useless. It does not have enough buffering to cope with
any real load. Using it on a news server is an especially bad
idea. Sun have some patches for the le driver for Solaris, but the
real fix is to fit a real ethernet card and stop using le0.
------------------------------------------------------------------------------
>From Marc Gibian
What time do your backups run? Or maybe there are some network intensive cron
jobs? A while back I ran into something similar... a number of systems would
drop off the netowrk in the middle of the night. It turned out this was being
caused by having too many ufsdumps running over a single shared ethernet (10
mbps). By staggering my ufsdump scripts, I reduced the traffic at any given
moment and my systems stopped dropping off the lan. I now use a higher quality
backup tool (Solstice Backup/aka Networker, though most will do), and since the
switch have not seen this problem.
------------------------------------------------------------------------------
>From Glenn Satchell
Are there also 'le0: memory error' messages in the console or
/var/adm/messages? I have come across this at a few different
sites now and th esolution has been to use a second ethernet,
le1 instead of the builtin port. Seems lik ethere's some sort of
load related hardware problem...
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:11:18 CDT