Thanks for the explanation. That does seem reasonable. We do go through a few hoops to make MPI work with Julia, but so long as it works fine for you - that is indeed great news!
-viral > On 14-Jun-2015, at 8:15 am, [email protected] wrote: > > My reason for using MPI is simply that I'm used to MPI, and t is very easy > for me to adapt existing Octave code that uses MPI to use Julia instead of > Octave. For code such as these Monte Carlo examples, I doubt that it would > make much difference if one were to use pmap, for example, instead of MPI. > For more complicated code that uses multiple sends and receives, the ability > to adapt existing MPI code is more important. I haven't done any performance > comparisons. So far, I find that Julia with MPI works very nicely with MPI, > though. > > On Sunday, June 14, 2015 at 11:59:46 AM UTC+2, Viral Shah wrote: > Unless you really need to use some MPI functionality, why not use the > built-in parallel processing capabilities? Does using MPI give a major > performance benefit in your case? > > -viral > > On Friday, June 12, 2015 at 3:38:34 AM UTC-4, [email protected] wrote: > https://github.com/mcreel/JuliaMPIMonteCarlo has code for doing Monte Carlo > using the MPI package (https://github.com/JuliaParallel/MPI.jl). A note > addressed to primarily to economists gives some discussion. The intention is > to get beginners started, but there is also a more complex example. Embedded > in that example, there is code for local linear nonparametric regression.
