Haai,

On Wed, September 12, 2007 01:13, Andrew Doran wrote:
> On Mon, Sep 10, 2007 at 02:06:25AM +0300, Mindaugas R. wrote:
>[snip]
> Affinity by PID seems odd,

Which reminds me, there should be a way to supply the kernel with a set of
rules ({PID,PGRP} (range), invoking luser, tty, executable path (if
applicable), procname, kernel process?, etc.) and have it use said rules
to affect the scheduling of new processes (CPU affinity, but also
niceties). This allows us nice high-level control in addition to the
low-level methods proposed earlier.

Personally, I'd use it to tame highly parallel environments (both hardware
and software (especially with lots of short-lived processes)) -- they tend
to get out of hand and confuse the hell out of the scheduler 8)

Baai,

De Zeurkous
-----------

Friggin' Machines!

>[snip]
>
> Thanks,
> Andrew
>