ddg_monomer: if MPI executable is reproducible and efficient?

Member Site Forums Rosetta 3 Rosetta 3 – Applications ddg_monomer: if MPI executable is reproducible and efficient?

Viewing 0 reply threads
  • Author
    • #2117

        Dear friends,
        It is my first time using an MPI executable to run the ddg_monomer. I tested it to mutate the first residue from D to A with different cores, which is compared to a serial job. The following is the results:

        Core number, time, ddg prediction
        1 (serial), 32 hours, -0.362
        2, 25.433 hours, -1.722
        4, 29.766 hours, 0.055
        8, 24.68 hours, 0.239
        24, 25.13 hours, -3.080

        So can I ask:
        1) As the ddg prediction vary a lot, is ddg_monomer reproducible?
        2) More cores seem to be not so efficient, why is that?
        3) I also tested pmut_scan_parallel.linuxgccrelease. When “-constant_seed” is removed, the outputs are always identical no matter whether “-jran” option is constant or not. So is pmut_scan_parallel better than ddg_monomer in terms of reproducibility?

        One of the option files for the MPI job as well as its job script .sh file are attached. I would appreciate it if someone can help me that. Thank you very much.

        Yours sincerely

    Viewing 0 reply threads
    • You must be logged in to reply to this topic.