OK Jeffrey thanks a lot, the answer was very helpful to me, at least I got to know the complexity of this issue.
--- On Wed, 4/4/12, Jeffrey Squyres <jsquy...@cisco.com> wrote: From: Jeffrey Squyres <jsquy...@cisco.com> Subject: Re: [OMPI users] Need Performance estimation of MPI_Bcast To: "Open MPI Users" <us...@open-mpi.org> List-Post: users@lists.open-mpi.org Date: Wednesday, April 4, 2012, 5:43 PM On Apr 4, 2012, at 9:56 AM, anas trad wrote: > Thanks Jeffrey, I understand, can you please guide me to such paper that give > more accurate estimation for the MPI_Bcast. I can't cite any specific papers (collectives are not my focus area, and I left academia several years ago); you'll need to do some literature searches. I know that there have been quite a few papers and research projects over the past 15+ years about algorithms for MPI collective operations (including bcast). A good starting place would probably be proceedings from Euro MPI (and its predecessor, Euro PVM/MPI). I'm guessing that there are probably also some papers/projects that do modeling of collective algorithms in heterogeneous latency environments (e.g., shared memory + network), but this is even farther off my focus area, so I don't even have any general areas to point to. Sorry! > > > --- On Wed, 4/4/12, Jeffrey Squyres <jsquy...@cisco.com> wrote: > > From: Jeffrey Squyres <jsquy...@cisco.com> > Subject: Re: [OMPI users] Need Performance estimation of MPI_Bcast > To: "Open MPI Users" <us...@open-mpi.org> > Date: Wednesday, April 4, 2012, 4:35 PM > > It's a bit more complicated than that; there are many factors involved: > > - how many peer MPI processes are involved in the broadcast > - how many of those peers are local (and will likely communicate via shared > memory, not a network device) and how many are remote > - what broadcast algorithm is used (there are many) > > So the simplistic logp model won't help you here -- you'll need to be more > complicated, and/or dig into the academic papers on these kinds of things. > > > On Apr 4, 2012, at 9:11 AM, anas trad wrote: > > > Hi Envoyé , > > > > Thanks for the reply but I want to estimate the time of executing MPI_Bcast > > function by mathmatics. I found that we can achieve this by using LogGP > > model with this equation: > > > > T = ceil(log2p) . (L + 2 · o + (n − 1) · G > > > > but that does not give good accuracy for time evaluation. > > > > > > > > --- On Wed, 4/4/12, Aurélien Bouteiller <boute...@eecs.utk.edu> wrote: > > > > From: Aurélien Bouteiller <boute...@eecs.utk.edu> > > Subject: Re: [OMPI users] Need Performance estimation of MPI_Bcast > > To: "Open MPI Users" <us...@open-mpi.org> > > Cc: "us...@open-mpi.org" <us...@open-mpi.org> > > Date: Wednesday, April 4, 2012, 3:50 PM > > > > You can use the intel mpi benchmark, or skampi. These two programs are > > designed to evaluate Moi performance. > > > > Envoyé de mon iPad > > > > Le 2012-04-04 à 08:46, anas trad <trada...@yahoo.com> a écrit : > > > >> Hi all, > >> > >> I need to know the time estimation of executing MPI_Bcast function on > >> Neolith Cluster. Please, can anyone show me how to achieve that. > >> > >> Thanks, > >> Anas > >> > >> _______________________________________________ > >> users mailing list > >> us...@open-mpi.org > >> http://www.open-mpi.org/mailman/listinfo.cgi/users > > > > -----Inline Attachment Follows----- > > > > _______________________________________________ > > users mailing list > > us...@open-mpi.org > > http://www.open-mpi.org/mailman/listinfo.cgi/users > > _______________________________________________ > > users mailing list > > us...@open-mpi.org > > http://www.open-mpi.org/mailman/listinfo.cgi/users > > > -- > Jeff Squyres > jsquy...@cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > > _______________________________________________ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users > _______________________________________________ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users -- Jeff Squyres jsquy...@cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ _______________________________________________ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users
