Alternative for SharedVariable

Question

I have the following code:

Answer = Table[Undefined, {1048576}];
SetSharedVariable[Answer];

ParallelDo[(

  eq = getEx[i];
  (* Some calculation here with eq *)
  Answer[[i]] = Solve[eq][[1]];

), {i, 1048576}];

Export["my_list.csv", Answer]

What getEx[i] does is to return me with an expression that I need to solve.

I need to get all 1M differents equations and save it´s answers, they are all independet of each other, so I could use parallel programming to speed thing a lot.

My problem is that with a SharedVariable, each kernel always goes to the main one for evaluation, so this code is almost slower then using a simple Do[] instead of ParallelDo[]

I would like to know how could I make this run better? Since I am just writing into a SharedVariable but never reading from that during the parallel programing.

Thanks to all!

Answer 1

Why not just use ParallelTable and dispense with the shared variable altogether? Also, make sure that the definition of getEx gets distributed: if it lives in Global that shouldn't be a problem, but if it lives in a package it might.

LaunchKernels[];
DistributeDefinitions[getEx];
solutions = ParallelTable[
  Solve[getEx[i]][[1]]
  {i, 1048576}
];
CloseKernels[];