Subject: Re: Load Measurement
To: NetBSD User's Discussion List <netbsd-users@NetBSD.org>
From: Mipam <firstname.lastname@example.org>
Date: 12/07/2004 12:37:54
On Mon, 6 Dec 2004, Greg A. Woods wrote:
> [ On Monday, December 6, 2004 at 13:32:47 (+0100), Martin Husemann wrote: ]
> > Subject: Re: Load Measurement
> > The "load" numbers don't realy mean anything objective. You can not compare
> > them between different operating systems. I've seen AIX systems at load 20
> > that were fine for interactive use, and other systems at load 2 that were
> > barely usable.
> Indeed and also one cannot really compare between architectures either.
> On my dual-400MHz-CPU AlphaServer 4000 the load can be up in the 20's
> and the system is still quite usable and smooth for interactive use.
> Meanwhile my 300Mhz P-II system, which for some common tasks is much
> faster than a single 400MHz alpha 21164a, can be crawling and jumpy with
> a load average even in the 10's.
> However they're both running the exact same NetBSD release. :-)
The responsiveness of a system also depends on the scheduler used.
Linux did a lot of work on the scheduler and FreeBSD as well with
introducing the ULE scheduler (not yet stable) to increase resonsiveness
even under high loads. So high loads don't mean bad news by default.