[gmx-users] timestep fraction complete

[Mark Abraham]

Jack Shultz wrote:
> I am hoping for an update every 10 minutes just to satisfy the anxieties 
> of volunteers crunching on my project, but every hour may be 
> satisfactory. And if I can do this without requiring any shell scripts 
> it would be better in my environment. So I started making some code 
> modifications which apparently did not work
>  
> src/mdlib/sim_util.c:93:void print_time(FILE *out,time_t start,int 
> step,t_inputrec *ir)
> src/mdlib/sim_util.c-94-{
> src/mdlib/sim_util.c-95-  static real time_per_step;
> src/mdlib/sim_util.c-96-  static time_t end;
> src/mdlib/sim_util.c-97-  time_t finish;
> src/mdlib/sim_util.c-98-  double dt, fc;
> src/mdlib/sim_util.c-99-  char buf[48];
> src/mdlib/sim_util.c-100-  FILE *progress="progress.txt";
> src/mdlib/sim_util.c-101-
> src/mdlib/sim_util.c-102-  if (!gmx_parallel_env)
> src/mdlib/sim_util.c-103-    fprintf(out,"\r");
> src/mdlib/sim_util.c-104-  fprintf(out,"step %d",step);
> src/mdlib/sim_util.c-105-  if ((step >= ir->nstlist)) {
> src/mdlib/sim_util.c-106-    if ((ir->nstlist == 0) || ((step % 
> ir->nstlist) == 0)) {
> src/mdlib/sim_util.c-107-      /* We have done a full cycle let's update 
> time_per_step */
> src/mdlib/sim_util.c-108-      end=time(NULL);
> src/mdlib/sim_util.c-109-      dt=difftime(end,start);
> src/mdlib/sim_util.c-110-      time_per_step=dt/(step - ir->init_step + 1);
> src/mdlib/sim_util.c-111-    }
> src/mdlib/sim_util.c-112-    dt=(ir->nsteps + ir->init_step - 
> step)*time_per_step;
> src/mdlib/sim_util.c-113-
> src/mdlib/sim_util.c-114-    if (dt >= 300) {   
> src/mdlib/sim_util.c-115-      finish = end+(time_t)dt;
> src/mdlib/sim_util.c-116-      sprintf(buf,"%s",ctime(&finish));
> src/mdlib/sim_util.c-117-      buf[strlen(buf)-1]='\0';
> src/mdlib/sim_util.c-118-      fprintf(out,", will finish %s",buf);
> src/mdlib/sim_util.c-119-    }
> src/mdlib/sim_util.c-120-    else
> src/mdlib/sim_util.c-121-      fprintf(out,", remaining runtime: %5d 
> s          ",(int)dt);
> src/mdlib/sim_util.c-122-      fc = (int)dt / ir->nstlist;
> src/mdlib/sim_util.c-123-      fprintf(stderr, "Fraction complete: %d 
> \n", fc);

You're doing integer division and writing an integer. That won't get you 
fraction complete, even if the ratio of dt (the remaining runtime in 
seconds) and ir->nstlist (the number of steps between neighboursearches) 
meant anything! Use

fprintf(stderr, "Fraction complete: %g\n", (step - ir->init_step) / 
(float) ir->nsteps);

Mark

> src/mdlib/sim_util.c-124-  }
> src/mdlib/sim_util.c-125-  if (gmx_parallel_env)
> src/mdlib/sim_util.c-126-    fprintf(out,"\n");
> src/mdlib/sim_util.c-127-
> src/mdlib/sim_util.c-128-  fflush(out);
> src/mdlib/sim_util.c-129-}
>  
> Back Off! I just backed up md.edr to ./#md.edr.4#
> starting mdrun 'Protein in water'
> 500 steps,      1.0 ps.
> step 100, remaining runtime:   122 s          Fraction complete: 0
> step 200, remaining runtime:    92 s          Fraction complete: 0
> 
> 
> On Thu, Sep 24, 2009 at 6:34 PM, Mark Abraham <Mark.Abraham at anu.edu.au 
> <mailto:Mark.Abraham at anu.edu.au>> wrote:
> 
>     Jack Shultz wrote:
> 
>         I figured out another way to update the progress of my
>         simulation, but I need to report the fraction of completion at
>         the certain intervals of mdrun. Possibly at every time step or
>         if that does not make sense every 100 timesteps. I don't think
>         this is a feature currently supported, so I will have to make
>         some source code chages. I am looking for the variables related
>         to total number of time steps and where it controls the the
>         current time step so I can calculate the fraction of complete
>         where current-timestep / total-timestep
> 
> 
>     Under at least some circumstances GROMACS writes the expected
>     runtime remaining to stderr, but I don't recall what/when. Piping
>     that to some useful script has to be a good start. You certainly
>     don't need this data every time step, and don't want to be
>     perturbing mdrun to get it.
> 
>     Even if the above was unsuitable, surely progress would only need to
>     be moderately accurate and for simulations that last many hours. If
>     so, you'll get good enough data by grepping a gmxdump of the .tpr to
>     get the expected frequencies of output and length of run, and then
>     watching the growth of whichever of .log/.trr/.edr is written most
>     frequently. I/O buffering will affect the numbers somewhat.
> 
>     Mark
> 
>     _______________________________________________
>     gmx-users mailing list    gmx-users at gromacs.org
>     <mailto:gmx-users at gromacs.org>
>     http://lists.gromacs.org/mailman/listinfo/gmx-users
>     Please search the archive at http://www.gromacs.org/search before
>     posting!
>     Please don't post (un)subscribe requests to the list. Use the www
>     interface or send it to gmx-users-request at gromacs.org
>     <mailto:gmx-users-request at gromacs.org>.
>     Can't post? Read http://www.gromacs.org/mailing_lists/users.php
> 
> 
> 
> 
> -- 
> Jack
> 
> http://drugdiscoveryathome.com
> http://hydrogenathome.org
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at http://www.gromacs.org/search before posting!
> Please don't post (un)subscribe requests to the list. Use the 
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php