Mailinglist Archive: opensuse (1620 mails)

< Previous Next >
Re: [opensuse] Re: Task & Resource Management (evolution? or NIH syndrome?)
On 2014-11-22 04:55, Linda Walsh wrote:

Another tool I use, rather than those, is "cpulimit". It forces the task
not to use more than a percent of the cpu, no matter how free the cpu is
or not. On a machine that is already running full time, it runs cooler.
Takes longer, so what? And if you need the cpu for other primary tasks,
you have it.[
That is one that will likely use more power.

If you mean watts, it is debatable, yes.

If you force a task to idle when
nothing else needs the processor, then your process will start losing
it's cache to other procs the longer it stays idle.

No, that is not the case.

A process is limited to, say, 50% or 80% or 90% of the available cpu
power, any figure you wish. The idea is that this intensive and long
task never goes beyond that figure, but otherwise takes as much as it
wants (till the limit), and never really idles, so its caches are there
— unless those primary tasks are also intensive.

It is curious that by choosing nice and a 90% limit, the desktop remains
very responsive and fast, because there is always some free cycles for
the jobs that watch for the user activity, instead of those processes
competing for cycles.


Control is not smooth. It is not done directly by the kernel, but via a
trick: the control process stopping and restarting the target task dunno
how many times per second. The CLI can be confused and think it went
into BG.

On modern CPUs, you use less electricity by having the task run as fast
as possible, then fully idling the cpu at the end, that by running it
slowly. I doubt this, specially if the cpu runs at a lower clock, but
there are apparently studies on this. I'm not sure about the fan and
cabinet heat control, though.

Cheers / Saludos,

Carlos E. R.
(from 13.1 x86_64 "Bottle" at Telcontar)

< Previous Next >
Follow Ups