I have to 2 questions about elapsed job runtimes. 

1. The same Scoring notebook is run 3 times as 3 Jobs.   The jobs are identical, same PetaStorm code, CPU cluster config ( not Spot cluster) and data but have varying elapsed runtimes.   Elapsed runtimes - Job 1:  3 hours,  Job 2 39 min, Job 3  42 min.      What would cause the large Job 1 variation?

1. The same Scoring code is ported to run on GPU cluster.  Updated environment and libraries were required.   Same data.  Runtimes: JobG 1 GPU single node cluster 67 min.  JobG 2  multi-node 1-4 cluster  73 min.   The JobG 2  multi-node did not run faster than JobG 1 single node.   Looking a logs, the nodes run serially and not parallel.  node 1 runs and stops, node 2 runs and stops, node 3 runs and stops, node 4 runs and stops.   Is there a configuration parameter that is required for parallel node processing?  Or another factor ?

Thanks for help.