Data set X291 with 3 EBs. Weblog is OFF for all tests.

casa-test-4.4.123, pipeline rev 32881, CV cluster

1. Each EB runs on a separated node with MPI, simultaneously. Each node runs on 14 cores.
   • X169d (cvpost036) --> 2015-04-02 13:42:06 ~ 2015-04-02 16:31:41 --> 2h 49min
     • log in: /lustre/naasc/scastro/work/pipeline/X291/X169d/X169d_14subms_casapy-20150402-134129.log
   • X66b (cvpost060) --> 2015-04-02 13:42:04 ~ 2015-04-02 16:36:28 --> 2h 54min
     • log in: /lustre/naasc/scastro/work/pipeline/X291/X66b/X66b_14subms_casapy-20150402-134134.log
   • X27f (cvpost023) --> 2015-04-02 13:42:03 ~ 2015-04-02 17:54:09 --> 4h 12min
     • log in: /lustre/naasc/scastro/work/pipeline/X291/X27f/X27f_14subms_casapy-20150402-134142.log
2. All 3 EBs run together on a single node with MPI. It uses 14 cores on this node.
   • PARALLEL: 2015-04-07 17:16:35 ~ 2015-04-08 01:42:48 --> 8h 30min
     • /lustre/naasc/scastro/work/pipeline/X291/parallel/X291_14subms_casapy-20150407-171539.log
   • SEQUENTIAL: 2015-04-08 09:21:37 ~ 2015-04-08 17:57:06 --> 8h 36min
     • /lustre/naasc/scastro/work/pipeline/X291/sequential/X291_sequential_casapy-20150408-092056.log

• Analysis of the results.
  • it’s much faster to run the EBs separated, if run simultaneously in different nodes. The worse case-scenario is dominated by the slowest node (50% faster). In the best case-scenario, it is about 67% faster than running the 3EBs together.
  • Each separated EB was run on a separated directory to avoid rewriting files from other EBs. It is possible for the pipeline to run all 3 EBs in separated nodes, in the same directory if the namespace is protected. Currently, the pipeline writes some files with universal names, independent of the MS name. Examples: byspw, flux.csv.