[ofa-general] Re: IPOIB NAPI

[Michael S. Tsirkin]

> Quoting Shirley Ma <xma at us.ibm.com>:
> Subject: Re: IPOIB NAPI
> 
> Michael,
> 
> >I have not benchmarked this, but actually the "return 1" version makes sense
> to
> >me too: since a new completion was observed after notify-cq, we likely
> currently
> >have HCA writing new completions into the CQ at a high rate, so it makes sense
> >to delay polling by a few cycles, and reduce the number of interrupts in this
> >way.
> 
> >Right?
> 
> Agree. Another question, have you benchmark IPoIB NAPI vs. missed event only
> mode: just change ipoib completion from notify-cq, poll-cq to poll-cq,
> notify-cq if any missed event, poll again? I am going to try this to see the
> performance difference.

At some point, I think I compared req notif + poll against
poll + req notif + poll if missed (both without NAPI), and did not see
any speed difference.

NAPI was also benchmarked and it was a win as compared to non-NAPI,
especially with multiple sockets tests.

-- 
MST