Re: resource limits on "interactive" processes

From: Paul David Fardy <pdf_at_porthos.ucs.mun.ca>
Date: Thu, 23 Mar 95 10:25:57 -0330

Your message dated: Wed, 22 Mar 95 13:42:54 MST
>
>Dear managers:
>
>I just installed the CERN-enhanced version of NQS 2.7 on our alpha-station.
>Per-process resource limits (including CPU time limits) are implemented and
>actually works...
>
>To encourage use of NQS batch queues it would be useful to be able to
>impose CPU limits on interactive processes and subprocesses...in a manner
>similar to interactive CPU limits on VMS. Can this be done under OSF/1?
>(clearly CPU limits are supported under OSF/1 since NQS can enforce it)...
>

Does NQS support per job CPU limits? For example

        set cpu_limit to 100 minutes

        submit job:
                forever {
                        system("my_big_model");
                }

If my_big_model runs for 99 minutes and the forever{system} loop
has less than a 1 second CPU overhead, this job could run for a year
and a half without exceeding per process CPU limits.

NQS can enforce per job limits by scanning the process tree and making
the parent job process responsible for its descendants. But then you
have the problem of monitoring orphans.

In the end, the only way to manage cheaters is with accounting.
Check the weekly logs to see that total CPU time used is consistant
with jobs submission. Give any minor offender a warning; give any
seriouse offender a boot to the head :-).

Or maybe none of _your_ users try to steal cycles.

        Paul Fardy

--
Paul David Fardy                     Memorial University of Newfoundland
Systems Programmer                   Elizabeth Avenue
Technical Support Group              St. John's, NF  A1C 5S7
Computing and Communications         CANADA
Received on Thu Mar 23 1995 - 08:56:51 NZST

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:45 NZDT