-- SandraCastro - 2014-11-13

1. Tests using MMS

We are running several tests on MMSs and will use this area to post some findings. None of the problems here are definitive and should not yet be reported. This is for our own reference. When the problems are confirmed, we will create appropriate JIRA tickets .

---

1. Cycle 2 data reduction script. Using mpi on two computers, 4 and 5 nodes.
   
   • the script stops in step 15 when running delmod on field='2'. Killing the casaplotms process releases the lock. This is reproducible. CAS-7139.

   • 2014-11-10 21:49:13 INFO delmod:::: Process 25438: waiting for write-lock on file /diskb/scastro/casa/work/split/MMS/uid___A002_X83bead_X515.ms.split/SUBMSS/uid___A002_X83bead_X515.ms.split.0006.ms/table.lock

   • The logger will start normally showing all the message in both computers and after awhile it stops showing messages.
2. Cycle 2 script, like in item 1 using 8 subMSs. Step 7, applycal on field='0' takes much longer than in sequential.
   
   • 8 mpi servers, it takes about 8 hours to run applycal on field='0'
3. Cycle 2 script, like in item 1 using 10 subMSs.
   • Commented out plotms to avoid the write-lock on the MS. The script runs through the end. Applycal this time is much faster, which is most likely due to the increased number of sub-MSs, so that the selected data is spread in more sub-MSs.
4. Cycle 2 script in sequential, using MS.
   • the script stops in step 13 when running gaincal. The task complaints of a write-lock not release somewhere. Killing a casaplotms which was left running from step 11 released the lock!
   • step 7, applycal takes about 30 minutes to run on field='0' in sequential.