On Thursday, February 9, 2012 4:52:15 AM UTC+1, Brent Millare wrote: > > From what I've seen, MPI in java is not good performance wise. You take a > huge hit due to all the copying. To get good cluster performance in java, > you need to use something like Java Fast Sockets. > http://developers.sun.com/solaris/articles/fastSockets.html > > http://www.des.udc.es/~juan/papers/comcom.pdf > > MPJ looks promising, but I honestly have never tried it. I'd be interested > in seeing how far that takes you. > > Best, > Brent > Dear Adam and Brent, Dear clojure users, We have been working quite actively in Java Fast Sockets and FastMPJ for the last 10 years and recently we have launched their webpages: http://www.fastmpj.com http://www.javafastsockets.com We are receiving a lot of attention since we made these webpages available, as well as from the our recent work, G.L Taboada, S. Ramos, R.R. Exposito, J. Tourino, R. Doallo, Java in the High Performance Computing arena: Research, practice and experience. In Science of Computer Programming <http://dx.doi.org/10.1016/j.scico.2011.06.002> (pre-print version <http://gac.des.udc.es/%7Egltaboada/papers/taboada_scp11preprint.pdf>), where we report about their exceptionally efficient support of Java communications on shared memory and low-latency Remote Direct Memory Access (RDMA) networks (e.g., InfiniBand). As you can see, MPJ can compete with MPI, considering the 20 years of code optimization of the MPI benchmarks compared to our internal MPJ benchmarks... MPJ results are great! Thus, FastMPJ achieve quite similar performance than MPI, sometimes even outperforms MPI (for shared memory transfers FastMPJ achieves 70 Gbps and Java Fast Sockets 50 Gbps, whereas MPI is below 40 Gbps). We have FastMPJ available in the downloads section and also we have gathered a quite nice documentation at http://fastmpj.com/documentation/ Regards, Guillermo -- Guillermo Lopez Taboada Dept. Electronics and Systems University of A Corunna, Spain http://gac.udc.es/~gltaboada -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en
