SLC (4/20) slows down over time...

[darrell at terra.ucsc.edu]

I have observed an interesting (!!) behavior with my SLC.  Over time
commands that do primarily I/O take longer to complete.  The delay can be
in executing the command, or in "finishing up" before I get the csh prompt
back.  The delay usually is in the range of 5 to 15 seconds.

First the configuration: 4/20, 104MB "lunch box", 16MB RAM, dataless
running 4.0.3c from a 4/390.  The 4/390 is virtually unloaded.  X11R4, but
also happens with SunView.

When I reboot the 4/20 commands complete quickly.  Over time (at few
hours), commands like "ls" that operate on something unlikely to be in the
cache take longer and longer to execute.

Perhaps "a long time to execute" is misleading.  They often seem to
execute OK, but the prompt from csh won't come back for a long time.  "ls
-la" on my home directory just took 11 seconds after printing the
directory before I got the prompt back.  I have 36 files in that directory
-- length of output does not seem to be a big factor.

"mail" is also a big offender -- after reading a message it may take up to
20 seconds before the prompt comes back, even on a small list of messages
(2 or 3).  After finishing "man rcsfile" the time to get the prompt was 9
seconds (I am experimenting as I type this).

I asked Sun and they said "Get the new boot PROM" -- the problem they
describe is a possible locking up of the CPU.  This is *not* the problem I
am seeing since all other windows remain operative.

This feels like an NFS problem.  However, I do not observe this on my
4/110 which gets service from the same server.  The difference is that the
4/110 boots from its local disk.

Has anyone else experienced this? Any ideas?  Suggestions?  Sun has not
come up with a solution (or the cause) yet.

Thanks, in advance, DL

Darrell D. E. Long
Baskin Center for Computer Engineering & Information Sciences
Applied Sciences Building
University of California
Santa Cruz, CA  95064

(408) 459-2616