Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 2 Feb 1996 15:36:13 +1100
From:      Bruce Evans <bde@zeta.org.au>
To:        jgreco@brasil.moneng.mei.com, peterb@telerama.lm.com
Cc:        andreas@knobel.gun.de, dennis@etinc.com, hackers@FreeBSD.org, hm@altona.hamburg.com, mrcpu@cdsnet.net
Subject:   Re: Multi-Port Async Cards
Message-ID:  <199602020436.PAA27552@godzilla.zeta.org.au>

next in thread | raw e-mail | index | archive | help
>> The observation that a FreeBSD box acting as a router is "mostly idle" is 
>> bogus; since routing takes place entirely in the kernel and "idleness" is 
>> a measurement of the number of processes in userland waiting to run, I'm 
>> not sure it's an accurate measure.

Wrong.  Idleness is an estimate of the amount of time the CPU spends idling.
It is quite accurate under FreeBSD.

>When comparing the execution times of a CPU intensive program as a measure
>of CPU loading, and vmstat's report of system and idle times, I have found
>that there is a strong correlation between how many packets the machine is
>routing per second and both of those numbers.

That's because FreeBSD accounts for most interrupt time fairly
accurately.  Other systems usually don't.  FreeBSD doesn't account for
serial interrupt time accurately.  This is probably important here.

>If I can run a CPU hungry program and it only takes 25% longer even though
>the box is routing hundreds of packets per second, I tell you the router is
>mostly idle.  The claim is certainly not bogus.

This is the only reliable test - run your applications and see if they
are affected too much by the load.

Bruce



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199602020436.PAA27552>