This is an interim update on my problem getting Amanda 2.2.6.6 to work on
my Alpha 3000/500 running OSF/1 3.2.
The problem is that Amanda starts up dump in sendsize-dump just long
enough to get the tape estimate, then it kills the process. dump, being
what it is, creates 3 semaphores and 1 shared memory segment that it does
not release when terminated this way, so it leaves cruft around. By
default OSF/1 3.2 allows a maxium of 16 semaphores per process (I think).
So after Amanda does 5 estimates, the table is full for the backup user
id. By adding the following lines to the kernel config file, rebuilding
the kernel and rebooting, I got enough room in the semaphore table to run
Amanda. Problem is, now I need to run a clean up script to remove all the
old semaphores before the next Amanda run starts. (hack, gag, cough).
Here are the mod's I put in my kernel config:
semmni 90
semmns 200
semmsl 60
Here's a URL with a good summary of the problem and info on what the
kernel parameters mean:
http://www.ornl.gov/cts/archives/mailing-lists/alpha-osf-managers/1994/10/msg00269.html
From: Phil Newman <newman_at_pandora.gsfc.nasa.gov>
suggested setting semmsl to 2500, but that seems (possibly) extreme. He
may be right though. I'll wait and see what tonights dump brings.
Thanks to all the people that offered help and to ORNL for archiving the
Alpha OSF Managers mailing list!
Have a Happy Thanksgiving!
Ken Mayer
Received on Thu Nov 23 1995 - 01:36:11 NZDT