Julia's MPI.jl is very similar to mpi4py. Both are thin wrappers around MPI, and should give the same performance as using MPI from C.
HPCC is a set of benchmarks for systems, not for languages. The algorithms used are prescribed, and using a high-level language has no chance of improving performance. This may test "abstraction penalty", i.e. whether one can write efficiently low-level code in a high-level language, but given that the benchmark implementations can be arbitrarily complicated and would not need to look like "well-written Julia" (whatever that means), I assume that Julia would show the same performance as C. An interesting benchmark would be to define a high-level non-trivial problem, such as e.g. "solve the Poisson equation on a grid with adaptive mesh refinement". Then, one can compare both speed and code complexity for different languages, and people could then choose according to their needs. Naively, I would expect Python to be good at code complexity, and C/C++ be good at speed, and even more naively I would hope that Julia is close to both of them... -erik On Wed, Oct 1, 2014 at 5:55 PM, Kapil Agarwal <kapil6...@gmail.com> wrote: > Hi > > What I understand is that comparing Julia and MPI may not be a good option > as both of them have different internal implementations. > > So Julia may be benchmarked against mpi4py and similar MPI ports ? Has such > benchmarking already been done ? If so, could you please share the results. > > Another thing that I was thinking was implementing the benchmark tests at > http://icl.cs.utk.edu/hpcc/ in Julia in parallel. Would that be a good > project and beneficial to the Julia community ? > > > Regards, > Kapil -- Erik Schnetter <schnet...@cct.lsu.edu> http://www.perimeterinstitute.ca/personal/eschnetter/
