How to utilize memory across multiple nodes on an HPC resource for an in-core solve

I am trying to run a very large model on our HPC resource, however, when I request multiple nodes that together have enough memory for the specified in-core required memory, it only recognizes a single node's memory and switches to out-of-core mode. Is it possible to utilize the memory across multiple nodes to achieve an in-core solve?


Thank you for the help,

Peter



Tagged:

Answers

  • mrifemrife PHLForum Coordinator

    Hi @PSchwarzenberg please double check the job submission. The 'Local memory...' is the amount needed by the master process: it reads in the input file; does the domain decomposition; then broadcasts the solution commands to the other cpu cores; and solves its own portion of the model. Since it is allocated close to 50% of the total memory (needed for all cpu cores used in solution including the master) it looks like only 2 cpu cores were requested to be used. Which I'm sure was not your intent.

    Perhaps discuss and show us how the job is being submitted.


    Mike

Sign In or Register to comment.