Performance monitoring (for the Navier–Stokes solvers)

    This logs simple statistics available for the various Navier–Stokes solvers.

    event perfs (i += 1) {
      static FILE * fp = fopen ("perfs", "w");
      if (i == 0)
        fprintf (fp,
    	     "t dt mgp.i mgp.nrelax mgpf.i mgpf.nrelax mgu.i mgu.nrelax "
    	     "grid->tn perf.t perf.speed npe\n");
      fprintf (fp, "%g %g %d %d %d %d %d %d %ld %g %g %d\n", 
    	   t, dt, mgp.i, mgp.nrelax, mgpf.i, mgpf.nrelax, mgu.i, mgu.nrelax,
    	   grid->tn, perf.t, perf.speed, npe());
      fflush (fp);

    If we have a display (and gnuplot works), a graph of the statistics is displayed and updated at regular intervals (10 seconds as defined in perfs.plot).

    event perf_plot (i = 10) {
      if (getenv ("DISPLAY"))
        popen ("gnuplot -e 'set term x11 noraise title perfs' "
    	   "$BASILISK/navier-stokes/perfs.plot "
    	   "& read dummy; kill $!", "w");

    The graph looks like this:

    Graph displayed by perfs.h

    Graph displayed by perfs.h

    Monitoring a distant run

    This can also be useful to monitor what is happening to a simulation running on a remote system. The script provides a convenient way to do this.

    Imagine you started a simulation including the perfs.h file on the “mesu” system at UPMC (typically using the queueing system) and that the simulation is running in the wave/level-10/ directory on mesu. You can then get a regularly-updated graph of what is happening using:

    bash $BASILISK/navier-stokes/

    See also