Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

From: Gleb Natapov (glebn_at_[hidden])
Date: 2007-06-27 10:06:17

On Tue, Jun 26, 2007 at 05:42:05PM -0400, George Bosilca wrote:
> Gleb,
> Simplifying the code and getting better performance is always a good
> approach (at least from my perspective). However, your patch still
> dispatch the messages over the BTLs in a round robin fashion, which
> doesn't look to me as the best approach. How about merging your patch
> and mine ? We will get a better data distribution and a better
> scheduling (on-demand based on the network load).
Attached patch adds this on top of my previous patch. The performance on
my setup is little bit worse with this patch applied.

> Btw, did you compare my patch with yours on your multi-NIC system ?
> With my patch on our system with 3 networks (2*1Gbs and one 100 Mbs)
> I'm close to 99% of the total bandwidth. I'll try to see what I get
> with yours.
Your patch SEGV on my setup. So can check and compare. I see this in
your patch:
+ reg = recvreq->req_rdma[bml_btl->btl_index].btl_reg;
But bml_btl->btl_index is not an index in req_rdma array and actually we
never initialize bml_btl->btl_index at all, so may be it would be a good
idea to remove this field at all. TCP never use reg so no problem there,
but for IB it should be valid.