Dear Experts,
I am trying to run the Hippocampal and brainstem MCR code on a cluster.
When I schedule a job with anywhere between 2-4GB of RAM per dataset (i.e., per core), I am getting both java and matlab errors (from java, it is a GC out of memory error, and from matlab it is a vague segmentation fault). If I schedule the job with 6+GB of RAM, things run just fine.
My sense is that there are maybe some multi-core/thread or memory settings in either java or the MCR that are causing trouble here? Or does this process really need 6GB of RAM?
If the RAM requirements are closer to the 2-3GB range, do you happen to know of any flags I can/should pass to java/MCR to limit the parallel/memory expectations? For example, I’ve tried: export _JAVA_OPTIONS="-XX:ParallelGCThreads=1”, but without any luck.
Let me know if you have any suggestions!
Al the best,
Ryan