> Thank you, Jeff and Ganesh.
> My current research is trying to rewrite some collective MPI
> operations to work with our system. Barrier is my first step, maybe I
> will have bcast and reduce in the future. I understand that some
> applications used too many unnecessary barriers. But here what I want
> is just an application to measure the performance improvement versus
> normal MPI_Barrier. And the improvement can only be measured if the
> barriers are executed many times. I have done some synthetic tests,
> all I need now are real applications.
Looks like a useful project.
BTW in our paper, we cite a paper by Rolf Rabenseifner. That paper is
definitely worth reading. It provides interesting performance studies
(the paper cited is old, so the studies are probably worth repeating).
Then again, in our paper we cite a paper by Siegel et al. That is also
definitely worth reading. In that paper, they mention a real-world
experience story of a barrier that was thought to be useless and was
removed... only to find, a year later, that it caused a communication
race (a wildcard receive now could obtain an extra match that was not
possible with the barrier present).
Good luck in your studies, Shan.
> users mailing list