[mpich-discuss] Why does MPICH perform busy wait?

Tim Jammer 3jammer at informatik.uni-hamburg.de
Sun Oct 2 06:02:06 CDT 2016


Thank you for your insights.

I am still wondering, why the busy wait is in the MPI_Barrier.
As the purpose of the Barrier is to just wait for all involved processes.

I think the latentency introduced with one or another way of saving 
energy is not such important in the Barrier.
For all other communication it might be best to perform a short busy 
wait in order to guarantee a maximum message rate.

Was it considered before to introduce an "energy aware wait" for the 
MPI_Barrier function?
So that the user can still enjoy the high performance solution for all 
other communication.

Tim Jammer

Am 29.09.2016 um 17:08 schrieb Rob Latham:
> Your proposal introduces latency.  Data centers want power efficiency, 
> but several large classes of applications care about message rate.

discuss mailing list     discuss at mpich.org
To manage subscription options or unsubscribe:

More information about the discuss mailing list