Skip to main content

Major difference between Linux (cluster) and Windows (computer) with same parameters

Answered

Comments

1 comment

  • Riley Clement
    • Gurobi Staff

    Hi Pauline,

    I can see in the first log that after the push phase there is a large primal infeasibility:

    Push phase complete: Pinf 3.7635916e+02, Dinf 4.1848068e-08    238s

    Then the simplex algorithm kicks in to try and clean up this infeasibility… and it's not performing great.  This is what is causing such a difference in timings.  In the second log there is no large infeasibility and we jump straight into the branch and bound tree.

    For some background it is probably worth reviewing the following two articles:

    * Why does Gurobi perform differently on different machines?

    * What is performance variability?

    So it is plausible that sometimes the Linux machine will perform well and sometimes it won't, and the same goes for Windows.

    To explore this properly you would need to run the model with several different values of Seed, on both machines, then compare the logs.  If you see the Linux machine consistently ending up with large infeasibilities after the push phase, and Windows doesn't, then perhaps it warrants further investigation.

    - Riley

     

     

     

    0

Please sign in to leave a comment.