cores number in parallel jemris

[Skander Salah]

Dear Tony and the Jemris developing team,

First let me express my gratitude and fascination as a student of biomedical computing for providing such a great tool. Thank you for your efforts.

I just managed to make parallel jemris work and I noticed that when I use a number of cores under 10, everything works fine and pjemris sends MR samples (100 spins in my test example from  jemris/angio_simu/simu  ) to up to 9 slaves without any problems but as soon as I choose a cores number higher than 10, the simulation reaches 101% (quite fast to be frank) and freezes there.

I double checked that I have 24 cores on my cpu (CPU(s): 24 when I use the command lscpu). I am on Linux 18.04 and I use openmpi v4.1.4. I ran my tests using the defualt example files in jemris/angio_simu/simu.

Is this a known limitation of the number of cores that can be used in pjemris? Or is there a way to use all the cores and further accelerate the simulations?

Thank you very much

Skander Salah

[Skander Salah]

I ran some further tests, and it seems to me that the maximum number of exploitable cores in pjemris is related to the number of spins in the MR sample: one simulation with 100 spins had a maximum of 10 cores, another simulation with 200 spins had a maximum of 20 cores. Another default brain sample with over 2000 spins can be simulated with all of my available 24 cores.

I feel I am missing some understanding of how things are working, can you please confirm this with a short explanation? or tell me how to overcome this limitation?
It would be actually a nice feature for pjemris in Matlab GUI, instead of the programmed default 4 cores,  to either have the number of cores as an input parameter (as long as the number of spins for each sample can be shown/known before the simulation is started) or to programmatically make the number of cores dependent on the number of spins if the user chooses to use all his available resources for the fastest simulation.

Thank you very much