On Tue, Nov 08, 2011 at 10:28:07PM -0800, Nicholas A. Bellinger wrote:
> Hi Christoph,
>
> What extra parameters do you recommend using for the alloc_workqueue()
> setup here..?
>
> While doing some initial fio large block performance tests with a 2x
> port 8 Gb/sec 25xx parts with this evening, I noticed the per HW port
> qla_tgt->qla_tgt_wq workqueues using alloc_workqueue("qla_tgt_wq,
> WQ_UNBOUND, 1); seem to have a slight performance edge over this patch
> to start using a single global qla_tgt_wq setup in qla_target.c with
> alloc_workqueue("qla_tgt_wq, 0, 0).
>
> It's on the order of a ~300 MB/sec difference between the two, with the
> per qla_tgt->qla_tgt_wq running very near what the backend is capable of
> at ~1500 MB/sec, and this patch to convert to a single qla_tgt_wq with
> the same tests are ~1200 MB/sec.
>
> For the slower case with this patch, a single kworker thread is running
> @ 100% CPU utilization with the wq defaults. Using per HW port context
> qla_tgt->qla_tgt_wq with WQ_UNBOUND, 1, multiple kworkers are running at
> ~40% utilization, and AFAICT seem to be doing a better job of
> distributing load across multiple qla_hw_data ports.
>
> Any thoughts on what might be limiting the single global qla_tgt_wq in
> this patch against per HW port qla_tgt_wq dispatch..?
Read through Documentation/workqueue.txt. It sounds like this workqueue
is CPU bound, although I wonder why. Can you look at perf top and see
were we spend the time?
--
To unsubscribe from this list: send the line "unsubscribe target-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
