Groups   126 of 99+      julia-users › SharedArray returns unexpected empty values 6 posts by 2 authors      Fred    Mar 26   Hi,  I did this simple code to try to understand how shared arrays work. Unfortunately my shared array c    is always empty :  Here I know the size of the final array c because it is a very simple example, but  is it possible to create an empty shared array and push the results inside using push   function ?  Thank you very much for your comments !  addprocs 2    a   rand 8   b   rand 8    convert SharedArray,a   convert SharedArray,b    c   SharedArray Float64,8     everywhere function prod x,y     z   x y    return z  end   parallel for i 1:10   c i    prod a i , b i   end  println c     result :  0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0  :           Yichao Yu    Mar 26   On Sat, Mar 26, 2016 at 3:46 PM, Fred  fred.so... gmail.com  wrote:    Hi,       I did this simple code to try to understand how shared arrays work.    Unfortunately my shared array c    is always empty :     Here I know the size of the final array c because it is a very simple    example, but  is it possible to create an empty shared array and push the    results inside using push   function ?       Thank you very much for your comments !       addprocs 2         a   rand 8      b   rand 8         convert SharedArray,a      convert SharedArray,b         c   SharedArray Float64,8          everywhere function prod x,y        z   x y       return z     end        parallel for i 1:10      c i    prod a i , b i      end   http:  julia.readthedocs.org en latest stdlib parallel  Base. sync   You also have a oob access btw.       Fred    Mar 26   Thank you for your answer !  I did the following change :   sync  parallel for i 1:8    c i    prod a i , b i   end  and now I obtain a result :  0.18576241230332016,0.23210296820071408,0.2741188848763749,0.6734257301922404, 0.4961438140080696,0.06445466347639998,0.07932590741374389,0.32153039018331936   I hope everything is correct :        Fred    Mar 26   I compared the speed of the parallel loop : it is 10 times slower  with 4 cpus  than the simple loop :    time  sync  parallel for i 1:800000    c i    prod a i , b i   end   println c    time for i 1:800000    c i    prod a i , b i   end  0.728079 seconds  391.50 k allocations: 16.659 MB  0.091947 seconds  4.80 M allocations: 73.240 MB, 5.67  gc time           Yichao Yu    Mar 26   On Sat, Mar 26, 2016 at 5:14 PM, Fred  fred.so... gmail.com  wrote:    I compared the speed of the parallel loop : it is 10 times slower  with 4    cpus  than the simple loop :     parallel use multiple processes and the synchronization is very high  compare to the super cheap operations in each loop.  Don't use it for simple code like this.   threads will likely be more suitable but there are a lot of work on  both frontend and backend for it to be more usable.       Fred    Mar 26   Thank you ! I will try on a more complex example ;