Distributed Simulations

General Usage
1

Hello! I’m trying to parallelize this jacobi code but it is not working, can someone help me, please?

using Distributed
addprocs(2)
@everywhere using LinearAlgebra
@everywhere using DistributedArrays
@everywhere  n=5
@everywhere  m=n^2
I=Diagonal(ones(n,n))
A = Tridiagonal([fill(-1, n-2); -1], fill(2, n), [-1; fill(-1, n-2);])
A=kron(A,I)+kron(I,A)
b=ones(m,1)
x=ones(m,1)
d=ones(m,1)
A=distribute(A; dist=(2,1))
b=distribute(b; dist=(2,1))
x=distribute(x; dist=(2,1))
d=distribute(d; dist=(2,1))
@everywhere maxit=10
#@everywhere dx=[]
diagon =Diagonal(A)
@sync begin
    for k in 1:maxit
      for i in 1:length(localindices[1])
         localpart[d][i]=diagon[i,i]
      end
       dx = (b-A*x)./d
       x = x+dx
       if norm(dx)< 0.00001
           break
       end
    end
end
x
2

I don’t know what the algorithm should do, but I see a few stuff that appears to be wrong in the @sync begin block. First, I think this block is only running in the master worker, you should use Distributed pmap, @spawn, etc to make the other workers actually work. Second, I think you are no using properly localindices and localpart, those are functions and you are trying to index it. Also diagon, if is intended to be part of the distributed part of the algorithm, must be defined @everywhere. Sorry I can’t help you in more detail, maybe if you isolate the independent issues unrelated of the real specific case, i can do more.
See PSA: make it easier to help you

1 Like