I am learning how to use Mathematica on a cluster, and a lot of this stuff is still way over my head. Please let me know if I am missing any crucial details.
The cluster uses Linux and uses Lmod to manage software modules. When I log into my account I then load Mathematica using the command
ml Mathematica/12.0.0, and then I start a Mathematica kernel using
math. Once the kernel is up, I put in the command
$MaxLicenseSubprocesses and I get back "Infinity".
What does "Infinity" mean here? If I run this on a notebook on my own computer it tells me 8 for
$MaxLicenseProcesses and 16
$MaxLicenseSubprocesses. I am assuming this corresponds to the largest number I can use in
LaunchKernels[n], since on my own computer if I try
LaunchKernels I get an error.
Does this mean I can use any number with the
LaunchKernels function? At what point does this become pointless in speeding things up? When I try parallel computations on my own computer, specifying
LaunchKernels vs the automatic
LaunchKernels which launches 2 kernels doesn't seem to create much improvement (as opposed to getting 16 kernels on the cluster which speeds up my calculations by a factor of 12).