dnaml_mon.c

fastDNAml is an attempt to solve the same problem as DNAML, but to do so faster and using less memo

/* dnaml_mon.c */

#define MAIN

#include  <sys/types.h>
#include  <stdlib.h>
#include  <unistd.h>
#include  <signal.h>
#include  <stdio.h>
#include  <string.h>
#include  <sys/times.h>
#include  <time.h>
#include  <math.h>
#include  "fastDNAml_types.h"
#include  "fastDNAml_funcs.h"
#include  "fastDNAml_globals.h"

#define  SIGINT    2

void    make_filename(char *, char *, char *, char *, int);
double  dwalltime00(void);
void    record_times(stat_data *sp);

/*  Global variables  */

#define MAXPROCS  1024
int          myprogtype = DNAML_MONITOR;
FILE         *fp, *outfp, *seqfp;
void         *p;
stat_data    xstat;
proc_data    proc[MAXPROCS];


/******************************************************************************/


int main(int argc, char *argv[])
  { /* main */
    int       type, from, sz, nextsp;
    char      filename[2048] = "";
    int       xtid;
    int       bufid;
    int       errno;
    char      c, *bp;
    char      buf[2048];
    double    step_time;
    double    ustime;
    double    twc;
    boolean   firstdone=TRUE;
    boolean   done;
    int       dnaml_ln, dnaml_cur;
    int       ntasks;
    stat_data *sd;

/* Initialize communication runtime library. */
    process_init(argc,argv,&myproc);

/* Resolve working directory path. Quit if there is an error. */
  if( realpath(workdir,buf) == NULL ) {
    fprintf(stderr,"%s: Bad path name: %s\n",argv[0],workdir);
    bail(argv[0],99);
  }
  strcpy(workdir,buf);
  strcat(workdir,"/");

# ifdef DEBUG
  make_filename(filename,workdir,run_id,".dnaml_mon.dbg",myproc.tid);
  if ((dbgfp = fopen(filename,"w")) == NULL) {
    bail(argv[0],ERR_DEBUGFILE);
  }
# endif

    make_filename(filename,workdir,run_id,".log",0);
    if((outfp = freopen(filename,"w",stdout)) == NULL) {
      bail(argv[0],ERR_LOGFILE);
    }
    fprintf(outfp, "Number of workers requested = %d\n", nworkers);
    nworkers = ntasks = 0;
    dnaml_ln = dnaml_cur = 1;


/*------------------------------------------------------------------------------
 * Now we enter a loop looking for messages from any process.
 */
    done = FALSE;
    while(!done) {
      from = ANY_SOURCE;
      type = ANY_TAG;
      probe_msg(&from,&type);

      switch(type) {

      case DNAML_INPUT_TIME:
        if(infol>0) {
          recv_msg(&xstat,1,from,type);
          step_time  = xstat.t1 - xstat.t0;
          twc = xstat.t1 - xstat.tstart;
          fprintf(outfp,  "file input : %8.3f  %11.3f sec\n", step_time, twc);
          fflush(outfp);
          fprintf(stderr, "file input done at %6.3f sec\n", step_time);
          if(infol>2) {
            fprintf(stderr, "   1\n   2\n   3 (%5d) ", DNAML_CHAR_COUNT*(dnaml_ln-1));
          }
        }
        nextsp = 4;
        break;

      case DNAML_TASK_ADDED:
        recv_msg(&proc[ntasks],1,from,type);
        if(proc[ntasks].progtype == DNAML_MASTER) {
          if(dnaml_cur>1) fprintf(stderr,"\n");
          fprintf(stderr,"+m %8.8x:%.32s\n",from,proc[ntasks].hostname);
          fprintf(outfp, "+m %8.8x:%.32s\n",from,proc[ntasks].hostname);
          dnaml_ln++;
          dnaml_cur = 1;
        }
        else if(proc[ntasks].progtype == DNAML_FOREMAN) {
          if(dnaml_cur>1) fprintf(stderr,"\n");
          fprintf(stderr,"+f %8.8x:%.32s\n",from,proc[ntasks].hostname);
          fprintf(outfp, "+f %8.8x:%.32s\n",from,proc[ntasks].hostname);
          dnaml_ln++;
          dnaml_cur = 1;
        }
        else if(proc[ntasks].progtype == DNAML_WORKER) {
          nworkers++;
          if(dnaml_cur>1) fprintf(stderr,"\n");
          fprintf(stderr,"+w %8.8x:%.32s   Total workers=%5d\n",
                   from,proc[ntasks].hostname,nworkers);
          fprintf(outfp, "+w %8.8x:%.32s   Total workers=%5d\n",
                   from,proc[ntasks].hostname,nworkers);
          dnaml_ln++;
          dnaml_cur = 1;
        }
        ntasks++;
        break;

      case DNAML_ADD_SPECS:
        recv_msg(&c,1,from,type);
        dnaml_ln = dnaml_cur = 1;
        fprintf(stderr, "\n%4d (%5d) ", nextsp++, DNAML_CHAR_COUNT*(dnaml_ln-1));
        break;

      case DNAML_SEND_TREE:
        recv_msg(&c,1,from,type);
        fputc('+', stderr);
        dnaml_cur++;
        break;

      case DNAML_RECV_TREE:
        recv_msg(&c,1,from,type);
        fputc('-', stderr);
        dnaml_cur++;
        break;

      case DNAML_STEP_TIME:
        recv_msg(&xstat,1,from,type);
        step_time  = xstat.t1 - xstat.t0;
        twc = xstat.t1 - xstat.tstart;
        fprintf(outfp, "~%4d Specs: %8.3f  %11.3f sec\n", nextsp-1,step_time,twc);
        fflush(outfp);
        fprintf(stderr, "\n%4d Step time = %8.3f ", nextsp-1,step_time);
        fprintf(stderr, "#### Total time = %11.3f seconds", twc);
        break;

      /* The following are used in cmon, but not currently in dnaml_mon. */
      case DNAML_WORKER_READY:
      case DNAML_AWOL:
        recv_msg(&xtid,1,from,type);
        break;

      case DNAML_DONE:
        recv_msg(&xstat,1,from,type);
        if(firstdone) {
          fprintf(stderr,"\n");
          firstdone=FALSE;
        }
        ustime = xstat.utime + xstat.stime;
        twc = xstat.t1 - xstat.tstart;
        fprintf(stderr,"-%8.8x:  u=%10.2f  s=%10.2f  u+s=%10.2f  wallclock=%10.2f\n",
            from, xstat.utime, xstat.stime, ustime, twc);
        fprintf(outfp, "-%8.8x:  u=%10.2f  s=%10.2f  u+s=%10.2f  wallclock=%10.2f\n",
            from, xstat.utime, xstat.stime, ustime, twc);
        nworkers--;
        if(nworkers+3 == 1) {
          done = TRUE;
        }
        break;

      }

      if (dnaml_cur % (DNAML_CHAR_COUNT + 1) == 0) {
        dnaml_ln++;
        dnaml_cur = 1;
        fprintf(stderr, "\n     (%5d) ", DNAML_CHAR_COUNT * (dnaml_ln-1));
      }
    }
/*------------------------------------------------------------------------------
 * Write out performance statistics for my process.
 */
    sd = &myproc.stats;
    record_times(sd);
    ustime = sd->utime + sd->stime;
    twc = sd->t1 - sd->tstart;
    fprintf(stderr,"-%8.8x:  u=%10.2f  s=%10.2f  u+s=%10.2f  wallclock=%10.2f\n",
        myproc.tid, sd->utime, sd->stime, ustime, twc);
    fprintf(outfp, "-%8.8x:  u=%10.2f  s=%10.2f  u+s=%10.2f  wallclock=%10.2f\n",
        myproc.tid, sd->utime, sd->stime, ustime, twc);
    fclose(outfp);

    bail((char*)NULL,0);
  } /* main */