Re: Linux Interrupt Latency benchmark v4 avaible -> detected irq bottleneck before do_bottom_half()
Alexander Kjeldaas (astor@guardian.no)
Mon, 31 Aug 1998 15:07:31 +0200
On Fri, Aug 28, 1998 at 03:29:44PM +0200, Andrea Arcangeli wrote:
>
> With an hacked version of lil + patch-ikd with kernel tracer enabled I
> exactly discovered what is stalling the irq. It' s the timer irq.
>
[...]
>
> So I think there' s really no way to reduce the maximal irq latency (that
> happens meanly _once_ after a continuous train of 200000 irq). I don' t
> think it' s a good idea to allow the timer interrupt to run with interrupt
> enabled.... A _maximal_ latency of ~2557 cycles on a 166Mhz P5noMMX
> (15,4uSec) is a very good time I think though... ;-). The mean is 5usec on
> the same hardware.
>
Hmm.. QNX guarantees 5us latency on a P133 so there _must_ be a way
for us to do better than 20us.
Btw, I ran your module (v3) for 14 hours on a PII233 and got 3.9us min
and 20.3max. The mean calculated by the module couldn't be trusted
because of the integer arithmetic, but from other tests with fewer
interrupts, I think it's very close to the minimum time.
astor
--
Alexander Kjeldaas, Guardian Networks AS, Trondheim, Norway
http://www.guardian.no/
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html