MATLAB: Parallel pool sizing for a parfor loop

parallel computingparfor

I run a parfor loop on an 8 cores workstation.

This loop is parfor i = 1:10 and the work in each iteration is substantial.

MATLAB automatically creates 8 workers.

Assuming that each iteration takes about the same time to complete, is this highly inefficient? I'm thinking that the 8 cores finish at about the same time, and then only 2 are assigned to the remaining 2 iterations.

Then again, I don't really understand the idea of parallel pools. How would I optimise the size?

Best Answer

Based on my own personal experience, the choice of the parpool size can be optimized only through experimentation. I have found that too many workers can be as counter-productive as too few, even when the selected pool size doesn't exceed the number of physical cores. There is an optimal middle ground that only experimentation allowed me to find.

Granted, this was a few years and a few MATLAB versions ago, and may have been a platform-dependent thing anyway... To be honest, I'm a bit surprised to hear that people still use parfor, given the time needed to launch a parpool. I use gpuArray all the time, but starting a parpool takes as long as most of the jobs I would like to accelerate.

Related Solutions

MATLAB: What happens when a parallel worker finishes early

When MATLAB runs a parfor loop, it attempts to split the iterations of the loop across the workers to keep them all busy simultaneously. However, this is not always possible (especially if you have only a very small number of loop iterations, and the iterations take different amounts of time). So, towards the end of the parfor loop, some of the workers will be idle waiting for the last workers to finish.

MATLAB: Use both multi core CPUs

You need to edit the 'local' parallel profile to set the maximum number of workers to 32. See the doc. Once you've done that, you should be able to issue the following command:

parpool('local', 32)

and launch 32 workers.

Note that if MATLAB by default does not consider hyper-threaded cores, as these tend not to perform well for numerically intensive computations. So, you might well find that running 32 workers is actually slower than running 16, depending on the precise nature of your computation.

Best Answer

Related Solutions

MATLAB: What happens when a parallel worker finishes early

MATLAB: Use both multi core CPUs

Related Question