Re: [patch 0/9] mutex subsystem, -V4

[Lee Revell]

On Mon, 2005-12-26 at 18:15 +0100, Mike Galbraith wrote:
> At 03:11 AM 12/26/2005 -0800, Andrew Morton wrote:
> >Arjan van de Ven <arjan@xxxxxxxxxxxxx> wrote:
> > >
> > >
> > > > hm.  16 CPUs hitting the same semaphore at great arrival rates.  The cost
> > > > of a short spin is much less than the cost of a sleep/wakeup.  The 
> > machine
> > > > was doing 100,000 - 200,000 context switches per second.
> > >
> > > interesting.. this might be a good indication that a "spin a bit first"
> > > mutex slowpath for some locks might be worth implementing...
> >
> >If we see a workload which is triggering such high context switch rates,
> >maybe.  But I don't think we've seen any such for a long time.
> 
> Hmm.  Is there a real workload where such a high context switch rate is 
> necessary?  Every time I've seen a high (100,000 - 200,000 is beyond absurd 
> on my little box, but...) context switch rate, it's been because something 
> sucked.

I can trivially produce 20K per second on my little sub Ghz box so 100K
on a busy server is certainly plausible.  Especially if for the purposes
of this discussion we are also worried about -rt + IRQ threading where
each IRQ costs two context switches (more if it raises a softirq).

Lee

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/