🚀 go-pugleaf

clangfor@mach1.wlu.ca (Carl Langford s) writes:

>We have a S81 running ptx v1.3. Once in a while, the system becomes quite
>sluggish. sar shows that the system disk (zd0) is 100% busy. The disk
>contains /, /usr and two swap partitions. sar also shows that there is no
>swapping on this system. /tmp is mounted as a separate partition on a 
>different disk. We have no user accounts on /usr, and, as far as we can
>tell, have no applications using / or /usr except for normal system
>activity. Can anyone provide some suggestions on what other things we should
>look for? The machine has been set up for timesharing (as opposed to
>commercial), has 14 processors and 144 meg of RAM. It typically supports 
>100+ users and the slowdown does not seem to be related to user load.

Swapping is not the only thing that would bring a system to its knees.
You might check the monitor program for excessive paging.  If this is
the case then your vmtune resident set size might be set too small.

You also might want to move one of you swap devices to another drive.  I
typically only have one swap device per spindle.

If you still have problems, you can also move /usr to a different drive.
I would bet that it is the paging problem though.

--
Robert Winter
Sequent Computer Systems
Costa Mesa, CA
winter@sequent.com

disclaimer: Opinions expressed here are my own.  Use at your own risk,
etc. etc.

In article <Bw2pG2.IAz@mach1.wlu.ca> clangfor@mach1.wlu.ca (Carl Langford s) writes:
>We have a S81 running ptx v1.3. Once in a while, the system becomes quite
>sluggish. sar shows that the system disk (zd0) is 100% busy. The disk
>contains /, /usr and two swap partitions. sar also shows that there is no
>swapping on this system. /tmp is mounted as a separate partition on a 
>different disk.

Nobody is shmallocing huge amounts of shared memory, are they?  I'm not sure if
this fits your problem, as shbrk() is likely to get stuff from elsewhere, and I
know nothing about ptx.  But on Dynix, shared memory may be allocated from
/usr/tmp if it can't get it from $TMPPATH, ./ or ~/ first.  It looks in
/usr/tmp before /tmp, according to shbrk(p).  It wouldn't show up as a file in
the file system, of course, you'd have to go poking around to find it.

Hope it's some help, Simon.
-- 
Simon Marshall, Dept. of Computer Science, University of Hull, Hull HU6 7RX, UK
  "Football isn't about life and death.  It's more important than that." Bill
Email: S.Marshall@Hull.ac.uk    Phone: +44 482 465181    Fax: 466666   Shankley

In article <5934@mccuts.uts.mcc.ac.uk> ehgasm2@uts.mcc.ac.uk (Simon Marshall) writes:
>In article <Bw2pG2.IAz@mach1.wlu.ca> clangfor@mach1.wlu.ca (Carl Langford s) writes:
>
>Nobody is shmallocing huge amounts of shared memory, are they?  I'm not sure if
>this fits your problem, as shbrk() is likely to get stuff from elsewhere, and I
>know nothing about ptx.  But on Dynix, shared memory may be allocated from
>/usr/tmp if it can't get it from $TMPPATH, ./ or ~/ first.  It looks in
>/usr/tmp before /tmp, according to shbrk(p).  It wouldn't show up as a file in
>the file system, of course, you'd have to go poking around to find it.
>
>Hope it's some help, Simon.

The model is different on PTX in that shared memory is not mapped through
the file system anymore. It is mapped directly into swap. /usr/tmp was 
hardcoded into the library routines that front calls to mmap() under 
Dynix 3. Under PTX they appear as 'real' system calls. Would suggest
evaluating paging activity, see vmtune(8).