C# Efficiently Dividing Tasks Between Cores

Question

I'm working on a small simulation that is running on my 8-core workstation. The simulation involves modeling the interactions between a large number of independent nodes. During one phase I need to perform a series of simple atomic operations to each node in parallel. I have been using Parallel.ForEach from System.Threading.Tasks to apply the operation concurrently to each node in the list of all nodes.

This worked well for the 100-500 nodes I used for testing. The load was balanced very well with all cores constantly utilized. Unfortunately, when I attempt to run the simulation with the main dataset (5000+ nodes), everything goes wrong. All 8 cores stay idle most of the time, spiking to 100% every few seconds and then returning to 1% utilization. After a few minutes of this an OutOfMemoryException is thrown and the program crashes.

I am not completely sure what is wrong, but remain suspicious that my current code is spawning many more threads than would be optimal for the task. I think the ideal method would be for the model to detect the number of available cores N, partition the list of nodes into N segments, then spawn N threads, giving each thread a separate partition of the list.

What I'd like to ask is if this is indeed a good solution to the problem, do better ones exist, and how should it be implemented in C#? Any advice or comments are welcome.

EDIT: Code sample by request

Parallel.ForEach(listOfNodes, tempNode =>
{
   tempNode.foo();
} );

<snip>

void foo()
{
   foreach(myType bar in listOfmyType)
   {
       if (bar.isActive)
           setNodeActive();
   }
} 

Answer 1

I think the ideal method would be for the model to detect the number of available cores N, partition the list of nodes into N segments, then spawn N threads, giving each thread a separate partition of the list.

Which is exactly what Parallel.ForEach does, so there must be another problem.

It's going to be very hard to come up with a better (Thread-management) system yourself. But you can use custom schedulers in the Task Library.

Answer 2

See this thread, which discusses limiting the number of threads that Parallel.For uses to avoid memory starvation:

http://connect.microsoft.com/VisualStudio/feedback/details/534571/parallel-foreach-may-create-an-inordinate-number-of-threads

I would try setting ParallelOptions.MaxDegreeOfParallelism to about 500, and see what happens.