On 05/07/2011 10:00 PM, Ingo Molnar wrote:
>
> * Asias He <asias.hejun@xxxxxxxxx> wrote:
>
>> BTW. The ping latency was caused by the movement of irq injection outside the
>> loop. Suppose we have 5 available buffers and only 1 buffer from tap device.
>> We will sleep on read without giving the buffer from tap to guest. The
>> latency will be huge in this case.
>>
>> while(virt_queue__available(vq)) {
>> ...
>> read(tap_fd)
>> ...
>> }
>> trigger_irq()
>
> But ... in one of the mails one of you claimed that even when moving the irq
> notification inside the loop (which we all agreed was necessary to avoid
> latencies!) the latencies would still occur during stress-tests.
What I got in stress-tests(ping -f) with irq notification inside the
loop is only network hangs.
>
> So something is still not understood here and could hit us anytime with any of
> the virtio drivers in the future and such bugs are not always so nice to debug
> like the latency problem here ...
Yes, Especially when they are under tremendous stress.
>
> Thanks,
>
> Ingo
>
--
Best Regards,
Asias He
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
