Performance question

mikea mikea at MIKEA.ATH.CX
Thu Oct 9 02:08:58 IST 2003


On Wed, Oct 08, 2003 at 04:51:30PM -0400, Lindsay Snider wrote:

> The best thing I would hope for would be 0% idle.  Then, you are processor
> bound, running the machine as fast as it can go.

It's OK for the idle to be 100% when there's no work to do, though,

Ideally, we'd have infinitely-fast, delayless I/O devices and infinite
quantities of delayless RAM. In that case, the work would arrive, the
CPU would run at 100% until the work was done, and then it could go
back to sleep.

In the real world, RAM is finite and has delays associated with it,
and devices have delays and finite bandwidth. Because of these, the
CPU runs in spurts as it gets enough data to work with, and waits
while the data are being read, fetched, and written. So instead of
_____________________________ __| |_______________

what *really* happens is

_|_|_||||||||||||_______|_|_|___||||___||||||___

over the interval in which CPU busy is being measured, because when
the CPU *is* busy, it's 100% busy, and when it's not busy, it's 100%
idle.

I'd love to have the infinite-bandwidth, delayless RAM and I/O. I
can make do with that and modern processors. If you can get me an
infinitely-fast processor, too, I can make you a *REALLY* fine filter.

--
Mike Andrews, waiting for the noumenon
mikea at mikea.ath.cx
Tired old sysadmin



More information about the MailScanner mailing list