On Apr 10, 2014, at 7:58 AM, Victor Vysotskiy <Victor.Vysotskiy_at_[hidden]> wrote:
> Dear Ralph,
>> it appears that 1.8 is much faster than 1.6.5 with the default settings, but slower when you set btl=tcp,self?
> Precisely. However, with the default settings both versions are much slower compared to other MPI distributions such as MPICH, MVAPICH, and proprietary ones.
Hmmm...odd, we aren't seeing that in our timing tests (running about even against MPICH, and against Cray when correct options are selected). Will have to explore that further.
> The 'btl=tcp,self' setting substantially boost performance of 1.6.5 on single processor multi-core computers running under Linux, or Mac OS X. But it doesn't work in 1.8 anymore ...
Doesn't make any sense - that option wouldn't have any impact whatsoever in the situation you describe with that code snippet. Something else must be going on...
>> Is this a copy/paste error?
> Unfortunately, it is not a copy/paste error.
>> the next question is: how many nodes were in your allocation?
> 2 processes on a single machine running under Ubuntu Linux (laptop), or Mac OS X (Mac mini).
Okay, I'll try to do a little poking around. Meantime, please send along the output from "ompi_info" so we can see how this was configured and what built.
> With best regards,
> users mailing list