Hello, in order to save time I would like to calculate a matrix function, let's say matrix(t), with different values of t for each kernel, the function remaining the same. A single matrix cannot be calculated in parallel. So the idea is to define a queue that is accessed by the kernels. When a kernel has finished the job, it writes the matrix on the disk and accesses the queue to get a t that is not yet being calculated by another kernel and calculates another matrix with this t. This goes on like this until a certain t is reached. I tried the following code with a reduced set of t's for testing:
ParallelEvaluate[
ClearAll["Global`*"];
SetDirectory["c:/functions"]; Get["matrix.m"];
SetDirectory["c:/matrices/"];
];
tValues = Range[100, 400, 100];
tQueue = CreateDataStructure["Queue", tValues];
Jobs[] := Module[{t, result},
While[Not[tQueue["EmptyQ"]],
t = tQueue["Pop"];
result = matrix[t];
Export[
"m_" <> ToString[t] <> "_ID_" <> ToString[$KernelID] <> ".txt", result, "Text"];
];
];
DistributeDefinitions[tValues, tQueue, Jobs]
parallelJobs = Table[ParallelSubmit[Jobs[]], {15}];
WaitAll[parallelJobs]
I tried different variants, also with ParallelEvaluate, but the result is always the same: the kernels calculate the matrices individually each for t = 100, 200, 300, 400 instead of sharing the calculation with the other kernels. What's wrong with this code?
