DistributedArray performance is pretty bad. The reason for removing them from base was to spur their development. All I can say at this time is that we are actively working on making their performance better.
For every parallel program you have implicit serial overhead (this is especially true with multiprocessing). The fraction of serial work to parallel work determines your potential parallel speedup. The parallel work / serial overhead in this case is really bad, so I don't think your observation is really surprising. If this is on a shared memory machine I would try using SharedArray's as the serial communication overhead will be lower, and the potential parallel speedup much higher. DistributedArrays only really make sense if they are in fact distributed over multiple machines. On Thursday, April 30, 2015 at 9:18:32 AM UTC-4, Alex wrote: > > Hi, > > I can't say anything regarding the performance of Distributed Arrays. > However, note that they have been relocated from Base to a separate > package: https://github.com/JuliaParallel/DistributedArrays.jl which > should work with 0.4-dev. > > Best, > > Alex. > > > On Thursday, 30 April 2015 12:10:47 UTC+2, Ángel de Vicente wrote: >> >> Hello all, >> >> I'm trying to understand the sort of performance that we can get in >> Parallel with Julia. DistributedArrays look very tempting, but my first try >> gives me a hopeless performance. As a test code, I got it from the slides >> (pages 75-80) at >> >> http://www.csd.uwo.ca/~moreno/cs2101a_moreno/Parallel_computing_with_Julia.pdf >> >> The code, which just defines two functions is available at: >> https://bitbucket.org/snippets/angelv/5kb4 >> and also attached to this message for convenience >> >> When I run it in my 8-core laptop in serial or in parallel (see below for >> output of the different runs), I see no better performance in parallel >> (though I see a huge increase in the allocated memory in parallel). With >> version 0.4-dev drand is not defined (well, actually the whole Distributed >> Arrays section is gone from the documentation for 0.4-dev). >> >> Any pointers on what can be done to improve this appreciated (this has to >> be the simplest possible parallel program, with no communication at all, so >> we should be able to get near perfect scalability here). >> >> Thanks a lot, >> Ángel de Vicente >> >> ========== >> >> angelv@pilas:~/mhdsolver-julia/Misc/Julia_Parallel$ julia -q >> julia> >> println(VERSION);require("simulation.jl");N=1000000;T=1000;A=rand(3,N);@time >> SimulationSerial(A,N,T) >> 0.3.7 >> elapsed time: 2.376680715 seconds (80 bytes allocated) >> >> ========== >> >> angelv@pilas:~/mhdsolver-julia/Misc/Julia_Parallel$ julia -p 4 -q >> julia> >> println(VERSION);require("simulation.jl");N=1000000;T=1000;dA=drand(3,N);@time >> >> SimulationParallel(dA,N,T) >> 0.3.7 >> elapsed time: 2.510426469 seconds (20011756 bytes allocated) >> 4-element Array{Any,1}: >> nothing >> nothing >> nothing >> nothing >> >> ============ >> >> angelv@pilas:~/mhdsolver-julia/Misc/Julia_Parallel$ >> /home/angelv/JULIA-DEV/julia/julia -q >> julia> >> println(VERSION);require("simulation.jl");N=1000000;T=1000;A=rand(3,N);@time >> SimulationSerial(A,N,T) >> 0.4.0-dev+4572 >> elapsed time: 2.33095253 seconds (283 kB allocated) >> >> ============= >> >> angelv@pilas:~/mhdsolver-julia/Misc/Julia_Parallel$ >> /home/angelv/JULIA-DEV/julia/julia -q -p 4 >> julia> >> println(VERSION);require("simulation.jl");N=1000000;T=1000;dA=drand(3,N);@time >> >> SimulationParallel(dA,N,T) >> 0.4.0-dev+4572 >> ERROR: UndefVarError: drand not defined >> >> >> >>
