[ofa-general] Re: [PATCH v7] IB/mlx4: shrinking WQE

[Michael S. Tsirkin]

> Quoting Roland Dreier <rdreier at cisco.com>:
> Subject: Re: [PATCH v7] IB/mlx4: shrinking WQE
> 
>  > ConnectX supports shrinking wqe, such that a single WR can include
>  > multiple units of wqe_shift.  This way, WRs can differ in size, and
>  > do not have to be a power of 2 in size, saving memory and speeding up
>  > send WR posting.
> 
> Given this added complexity:
> 
>  6 files changed, 226 insertions(+), 39 deletions(-)
> 
> and the unpleasantness of having if (BITS_PER_LONG == 64) various
> places,

I don't there's a way around that.
BTW, the vmap trick is an improvement in itself,
we can extend it to CQs, EQs etc easily.

> can you quantify the improvement this gives?

This gets me from 960 to 1020 MByte/sec on ipoib/cm with netperf.
SDP shows similiar gains.

> Would it make more sense to do this for userspace first?

Given that we want it, what does a delay buy us?

-- 
MST