Path: utzoo!attcan!uunet!wuarchive!zaphod.mps.ohio-state.edu!uakari.primate.wisc.edu!dogie.macc.wisc.edu!uwvax!ordin.cs.wisc.edu!hafner
From: hafner@ordin.cs.wisc.edu (Brian J. Hafner)
Newsgroups: comp.sys.sequent
Subject: Re: Timing parallel programs on Balance
Message-ID: <10705@spool.cs.wisc.edu>
Date: 28 Jun 90 03:21:18 GMT
References: <63900005@uxa.cso.uiuc.edu>
Sender: news@spool.cs.wisc.edu
Organization: U of Wisconsin CS Dept
Lines: 31

In article <63900005@uxa.cso.uiuc.edu> job00542@uxa.cso.uiuc.edu writes:
>
>I am working on a parallel Fortran program for the Sequent Balance and am 
>having difficulties getting accurate timings because the results vary with the 
>system load, even though I'm measuring CPU time.  I've heard that gang 
>scheduling of processes will help-- does anyone know how to do this?  Thanks in advance...
>
>James Bordner

Gang scheduling is a way to utilize a group of processors without
any interference from the OS.  Essentially, the processors are removed
from the "pool" of free processors and completely assigned to the
user.

I knew that the Encore Multimax could do this, but wasn't aware that
the Sequent [Balance | Symmetry] could.

Sequent offers two privledged system calls:

	proc_ctl() which allows a
	user to fix the priority of a process and prevent aging.

	tmp_affinity() which glues a process to a particular processor.

Although the OS still is charge of all the processors, these calls
can be used to fake gang scheduling.

Brian J. Hafner
Computer Sciences Department
University of Wisconsin - Madison
hafner@cs.wisc.edu