[openib-general] Re: How about ib_send_page() ?

[Michael S. Tsirkin]

Quoting r. Roland Dreier <roland at topspin.com>:
> Subject: Re: How about ib_send_page() ?
> 
>     Jeff> (side note: it would seem IPoIB could be re-written to
>     Jeff> dramatically improve it's performance).
> 
> Out of curiousity, what would the rewrite change to obtain better
> performance?
> 
> Thanks,
>   Roland

I have some ideas worth investigating:

1. I think an option to enable zero copy would help for the send side.
   This means we have to set NETIF_F_NO_CSUM so it likely cant be done
   with current hardware without breaking IPoverIB spec compliance,
   but we dont need a spec just to write code.

2. I dont think NAPI was ever tried on gen2 code. Worth a try?
   Might help more in combination with (1).

3. There's a draft for connected mode. That would increase the MTU
   and so help the receive side as well.


-- 
MST - Michael S. Tsirkin