[2614] in SIPB_Linux_Development
please don't access sipb-nfs with 2.1/2.2 kernels
daemon@ATHENA.MIT.EDU (mhpower@MIT.EDU)
Sun Feb 14 03:16:58 1999
From: mhpower@MIT.EDU
Date: Sun, 14 Feb 1999 03:16:46 -0500
To: linux-dev@MIT.EDU, gjking@MIT.EDU
Over the past few days the server sipb-nfs.mit.edu (canonical name
bloom-picayune.mit.edu) has been inaccessible often due to it wedging
or crashing. Recently, IP traffic to the machine has been monitored
regularly, and it appears that the wedge/crash instances occur during
times when files under /redhat are being accessed via NFS by Linux
machines using kernels more recent than the 2.0.x series, in
particular equal-rites.mit.edu and cliffs-of-insanity.mit.edu. Also,
the problem is reproducible in that generating heavy NFS traffic from
equal-rites.mit.edu will reliably either wedge or crash
bloom-picayune, whereas with correspondingly heavy traffic from Linux
2.0.35 machines (using the standard NFS client code supplied by Red
Hat), and from Solaris 2.6 machines, bloom-picayune continues running
normally. So far, it seems that NFS readdir operations are the most
likely source of the problem, but other NFS traffic may be problematic
as well. Another aspect of this is that listing directories via NFS
from equal-rites.mit.edu is unreliable -- not all of the files are
listed, and the number listed can vary from time to time.
It's not yet clear exactly what should be done to address this issue,
but for now please don't access any files on sipb-nfs via NFS from
Linux machines with kernels more recent than the 2.0.x series.
(The information on kernel versions from a few of the Linux machines
that had recently mounted directories under /redhat was found by
accessing port 79, 25, or 22 on these machines from charon2.mit.edu,
using "nmap -O".)
Matt
