Still trying to isolate cause.

   (1) Rerun PE test.
       (a) Also run top on compute node, mseas-data during
           "stalled" write.  Look for memory leaks (high swap)
       (b) Get strace and profiling software/information from
           Constantinos.  Run PE test with these.
           Need to run each in three configurations
              (i)   writing to local disk
              (ii)  writing to /home area
              (iii) writing to /gdata area

   (2) Try to reproduce slowness outside of PE model.
       Write FORTRAN code to write big array
          (a) in native binary
          (b) in netCDF
       Need to run each in three configurations
          (i)   writing to local disk
          (ii)  writing to /home area
          (iii) writing to /gdata area
       See write_test_job2

   (3) Isolate PE test as a package.  Give to Constantinos to
       test on another gluster system.