Date: Mon, 21 Feb 2005 22:02:53 -0700 (MST) From: "M. Warner Losh" <imp@bsdimp.com> To: das@freebsd.org Cc: current@freebsd.org Subject: Re: Load over 1000 Message-ID: <20050221.220253.125093887.imp@bsdimp.com> In-Reply-To: <20050221223518.GA25518@VARK.MIT.EDU> References: <45820.1109020342@critter.freebsd.dk> <20050221213337.GC87259@opteron.dglawrence.com> <20050221223518.GA25518@VARK.MIT.EDU>
next in thread | previous in thread | raw e-mail | index | archive | help
In message: <20050221223518.GA25518@VARK.MIT.EDU>
David Schultz <das@freebsd.org> writes:
: On Mon, Feb 21, 2005, David G. Lawrence wrote:
: > > No, disk I/O sleeps is not involved.
: > >
: > > The loadavg is the length of the runqueue. Any process sleeping,
: > > on network, disk or timer, is not counted towards the total.
: >
: > I said "historically". :-)
: > This was changed in FreeBSD a some years ago.
:
: Even further back in history, TENEX computed the load average
: based on runnable jobs. :-P See footnote 1 of RFC 546.
I was once told, but never went and looked at the source, that the
load average on TOPS-20 was computed as 'the number of milliseconds a
job had to wait for 1 millisecond of time'. The TOPS-20 scheduler let
you do things like say 'this class of people get all the time they
want' and 'that class of people can use no more than 10% of the CPU
when the load is over 3' and the like. So the different classes might
only have one or two jobs in the run queue, but still have a load
average of 125! Of course, if you were in one of the bad classes,
then it could take 15 minutes to get enough time to type 'uptime'.
But this is getting a little far afield...
Warner
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050221.220253.125093887.imp>