[openib-general] Re: [PATCH] IPoIB splitting CQ, increase both send/recv poll NUM_WC & interval

[Shirley Ma]

Leonid Arsh <leonida at voltaire.com> wrote on 04/27/2006 01:24:49 AM:

> Shirley Ma wrote:
> > Without seeing your patch, I coudn't say anything. I guess your 
> > implemention
> > didn't handler multithreads simultanously. If you only have one 
> > interrupt handler,
> > couldn't see any reason you can get better performance number with
> > splitting CQs.
> Shirley, you are right.
> I just wanted share our experience with you.
> 
> All the tests we made on our IPoIB driver, so our NAPI implementation 
> isn't relevant here.
> Unfortunately, we didn't plan to work on the IPoIB performance in the 
> nearest future, so I can't
> implement NAPI on the OpenIB driver right now.
> 
> I think it would be very interesting to compare the NAPI performance 
> against the work queue.
> Please let me know if you are planning to do it yourself. 
> >
> > Could you please post your NAPI patch here?
> >
> > As I mentioned I will test my patch to see how's the performance.
> 
> 
> Thank you,
>    Leonid

How many percentage throughput you got from your NAPI implementation?

So far work queue gives very consistent 15% througput increase in my 
local test with one dual core cpu over mthca. I am planning to add one 
more 
cpu to see the difference.

Yes, NAPI is in our plan. We can see NAPI vs. work queue results soon.
Actualy if we can combine work queue with NAPI, that would be more
interesting.

Thanks
Shirley Ma
IBM Linux Technology Center
15300 SW Koll Parkway
Beaverton, OR 97006-6063
Phone(Fax): (503) 578-7638

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20060427/b0f4fe72/attachment.html>