Re: safe load average
On Tue, May 07, 2002 at 08:17:11AM -1000, Joseph Dane wrote:
> >>>>> "Kirk" == Kirk Strauser <firstname.lastname@example.org> writes:
> Kirk> "Safe"? I think you should be more interested in CPU states
> Kirk> than load average. For example, consider running 50 webserver
> Kirk> processes, all of which are in an I/O wait state. Your load
> Kirk> average may be near 50, but your CPU may be sitting mostly
> Kirk> idle. As another example, pretend you're running an RC5 or
> no. "load average" refers to the average length of the run queue.
> processes blocked on I/O are not on that queue.
The load average refers to the average number of processes that are
runnable or in uninterruptible sleep. The latter usually indicates I/O.
(See kernel/timer.c:count_active_tasks() (in 2.4, at least) and ps(1).)
Colin Watson [email@example.com]
To UNSUBSCRIBE, email to firstname.lastname@example.org
with a subject of "unsubscribe". Trouble? Contact email@example.com