From owner-freebsd-hackers  Sun Nov  3 03:03:20 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id DAA14884
          for hackers-outgoing; Sun, 3 Nov 1996 03:03:20 -0800 (PST)
Received: from cheops.anu.edu.au (avalon@cheops.anu.edu.au [150.203.76.24])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id DAA14877
          for <hackers@freebsd.org>; Sun, 3 Nov 1996 03:03:15 -0800 (PST)
Message-Id: <199611031103.DAA14877@freefall.freebsd.org>
Received: by cheops.anu.edu.au
	(1.37.109.16/16.2) id AA042758993; Sun, 3 Nov 1996 22:03:13 +1100
From: Darren Reed <avalon@coombs.anu.edu.au>
Subject: Re: "Thrashing" index
To: julian@whistle.com (Julian Elischer)
Date: Sun, 3 Nov 1996 22:03:13 +1100 (EDT)
Cc: hackers@freebsd.org
In-Reply-To: <3278025C.41C67EA6@whistle.com> from "Julian Elischer" at Oct 30, 96 05:35:24 pm
X-Mailer: ELM [version 2.4 PL23]
Content-Type: text
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

In some mail from Julian Elischer, sie said:
> 
> Does anyone have any ideas on how to describe or detirmine the 
> amount of thrashing going on on a machine?
> 
> one thing I can think of would be to count the number of processes
> sitting in 'vmwait' state, or to look at teh page-in and page-out
> numbers, but hte problem with the second approach is that
> you need to look twice separated by a time period, to detirmine if the
> system is doing a lot of work and should not be asked to do more...
> 
> basically I'm trying to make the system self limit when memory starts
> to become in short supply, in combination with a few other events.
> 
> for example.. If the cpu idle time is high, but there is not much
> memory free, and there are processes in vmwait, then it's probably
> not a good idea to launch more processes, as the system is probably
> thrashing..
> 
> another index might be to find out how long it takes to allocate an 8K
> region or similar..
> 
> but  I think I'd like to find some way that you can look at (say) 
> 3 or 4 static indeces instantly available (e.g. through /proc or sysctl)
> and munge them into a go/no-go decision for launching more work.

So how do you, as the sys admin login or if logged in do anything about it ?