There is a simple (and good) reason not to have it in the upcoming
release. By lack of time, I didn't manage to test it well enough. As
I'm not 100% confident that it will not create any problems, I
preferred to left it out of the 1.3.
If you want to test it, please fell free to do so. And let me know if
you have any problems.
george.
On Oct 10, 2008, at 6:03 PM, Eugene Loh wrote:
Can someone tell me about mca_btl_sm_sendi()? In btl_sm.c, I see
that it's commented out of the structure "mca_btl_sm".
It seems to me that pingpong latency over shared memory in OMPI
isn't as fast as certain "competitors". If I put mca_btl_sm_sendi
back in, it seems to improve the pingpong latency a little. For
some pingpong test and some hardware and some compiler, etc., etc.,
I get 953 nsec out of the box and 902 nsec if I use
mca_btl_sm_sendi. So, it seems to improve latency.
Why is it commented out? E.g., look at btl_sm.c and search for the
first occurrence of "sendi":
mca_btl_sm_t mca_btl_sm = {
{
&mca_btl_sm_component.super,
...
NULL /*mca_btl_sm_sendi*/, /* send immediate */
...
}
};
(I'm just about to leave for a week, but I look forward to reading
everyone's insightful comments and lively discussion upon my return.)
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
