Summary: OSF/1 3.0 memory usage is driving me crazy

From: Perry Engle <perry_at_netplaza.com>
Date: Wed, 13 Mar 1996 10:45:35 -0500

Many Thanks to all who answered:

alan_at_nabeth.cxo.dec.com, Jim Belonis, Larry Church, Anthony
baxter,Patrick O'Brien,
Helgi Viggosson, SSTROBEL10_at_aol.com

here is a summary of their responses, that I did, and/or looked at:

Remove /sbin/swapdefault and use lazy swap
[ this was the most frequent response - it made tons of sense, and I'd
like to
increase my swap space ASAP. Until then, I changed to lazy swap, and
things didn't
seem to imporve, or even change noticably ]

For page/swap space usage you don't want to look at
        the RSS values, but the VSZ (virtual size). This
        is especially true you're using eager page/swap
        space allocation.

[ I've been looking at VSZ, it is usually pretty low, too, but I'll look
more closely]

Also monitor over-all swap space with
swapon -s

A process may fail to start because there isn't enough
        page/swap space for the required virtual memory. Enough
        real will be found by making other processes lose pages.
        A process may fail to start becuase the virtual memory
        requirements exceed some maximum.

To see the per-process limits use the C-shell built-in
        "limit". The system management guide will explain how
        to raise these limits. For most things the system wide
        limit on virtual memory is merely constrained by the
        amount of page/swap space available, but there may be
        configuration file or sysconfigtab limits.

To determine where your problem is, it would help to
see how the memory is divided between active, inactive, wired and
free list as well as the output from vmstat -M, ipcs -ma (shared memory),
dbx -k /vmunix -> print vm_perfsum, sysconfig -q vm and ps auxw.

[ In dbx -k /vmunix, I get the error:
"warning: Files compiled -g3: parameter values probably wrong."
I recompiled the kernal, and made sure that the -g flag was set,
 I have the script file to prove it, and I still get the error.
The vpf_ubcalloc is 33810 which seems way too low, but I'm stuck.
Based on what I'm seeing, I'm guessing that the UBC or something similar
is still my problem ]

[ipcs -ma gives this error message:
/dev/kmem: No such device or address]

[sysconfig is not on my system - is this a separate product?]

If you are running on the default vm parameters, the system tries
to keep the free list 128 pages or larger, and then dynamically divides
the rest between the UBC and VM (e.g. if you are not requiring to much
VM the UBC will become fat) You then are only having a problem,
if the free list is often dropping below 128 and as a consequence
experiencing
to much paging activity, hurting the performance of the system.
 
[ Here is what I have yet to do: ]

As I vaguely recall, OSF/1 3.0 had som "interesting" memory leaks. I
think
there were some patches to fix some of that. Upgrading to 3.2 is the best
advice. OSF/1 3.2b (it became DEC Unix around 3.2c or d) has been like a
rock for me.

boost the maxusers number
in the kernel config file, rebuild the kernel, and reboot.
 
[ my maxusers is 32, and I seldom have more than 5 or 6 users on the
system, including daemons ]

[ Most of all - thanks for the help, it is awesome - I think I'm way over
my head, and I'll look into some system administration classes at
Digital. Until then, I'll try to track down someone who can help me
out. Anyone know someone who can do some consulting for us? ]

Perry Engle (perry_at_netplaza.com)
RMI Marketing Inc, Hampton NH
Voice 603-926-6233
Received on Wed Mar 13 1996 - 17:28:33 NZDT

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:46 NZDT