Parallelization problems

[hildebra]

Dear Richard, thanks for providing this useful package to R, I really like working with it. However, recently I tried to upscale my analysis to 1000's of samples and needed to move to our Linux cluster. Here, the multithreaded jobs are causing problems: as far as I can infer this, I think the threads are finished, but are not being collected by the main thread, and this seems to happen random, or maybe it's related to the time single jobs take, as for small examples multicore works. Basically I see mostly on the stdout in verbose mode this being the last line: Runs: |================================================ | 95% . On the cluster where this job runs, the processes (all 20 in cases of 20 cores) are at 0% CPU usage and just doing .. nothing. Note that I'm using a miniConda R distribution (3.4.3) and the latest stable version of NMF from CRAN.

thanks a lot for any feedback, Falk