mpi-io-test.c 12.9 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443
/*
 * (C) 1995-2001 Clemson University and Argonne National Laboratory.
 *
 * See COPYING in top-level directory.
 */

/* mpi-io-test.c
 *
 * This is derived from code given to me by Rajeev Thakur.  Dunno where
 * it originated.
 *
 * It's purpose is to produce aggregate bandwidth numbers for varying
 * block sizes, number of processors, an number of iterations.
 *
 * This is strictly an MPI program - it is used to test the MPI I/O
 * functionality implemented by Romio.
 *
 * Compiling is usually easiest with something like:
 * mpicc -Wall -Wstrict-prototypes mpi-io-test.c -o mpi-io-test
 *
 * NOTE: This code assumes that all command line arguments make it out to all
 * the processes that make up the parallel job, which isn't always the case.
 * So if it doesn't work on some platform, that might be why.
 */

#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <unistd.h>
#include <string.h>
#include <sys/time.h>
#include <mpi.h>
#include <errno.h>
#include <getopt.h>

/* DEFAULT VALUES FOR OPTIONS */
static int64_t opt_block     = 16*1024*1024;
static int     opt_iter      = 1;
static int     opt_coll      = 0;
static int     opt_correct   = 0;
static int     opt_sync      = 0;
static int     opt_single    = 0;
static int     opt_verbose   = 0;
static int     opt_rdonly    = 0;
static int     opt_wronly    = 0;
static char    opt_file[256] = "test.out";
static char    opt_pvfs2tab[256] = "notset";
static int     opt_pvfstab_set = 0;

/* function prototypes */
static int parse_args(int argc, char **argv);
static void usage(void);
static void handle_error(int errcode, char *str);

/* global vars */
static int mynod = 0;
static int nprocs = 1;

int main(int argc, char **argv)
{
   char *buf, *tmp=NULL, *check;
   int i, j, v, err, sync_err=0, my_correct = 1, correct, myerrno;
   double stim, etim;
   double write_tim = 0;
   double read_tim = 0;
   double read_bw, write_bw;
   double max_read_tim, max_write_tim;
   double min_read_tim, min_write_tim;
   double ave_read_tim, ave_write_tim;
   double sum_read_tim, sum_write_tim;
   double sq_write_tim, sq_read_tim;
   double sumsq_write_tim, sumsq_read_tim;
   double var_read_tim, var_write_tim;
   int64_t iter_jump = 0;
   int64_t seek_position = 0;
   MPI_File fh;
   MPI_Status status;
	MPI_Comm comm;
   int nchars=0;
   int namelen;
   char processor_name[MPI_MAX_PROCESSOR_NAME];

   /* startup MPI and determine the rank of this process */
   MPI_Init(&argc,&argv);
   MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
   MPI_Comm_rank(MPI_COMM_WORLD, &mynod);
   MPI_Get_processor_name(processor_name, &namelen); 
   
   /* parse the command line arguments */
   parse_args(argc, argv);

   if (opt_verbose) fprintf(stdout,"Process %d of %d is on %s\n",
									 mynod, nprocs, processor_name);

   if (mynod == 0) printf("# Using mpi-io calls.\n");

    //sleep(15);
   
   /* kindof a weird hack- if the location of the pvfstab file was 
    * specified on the command line, then spit out this location into
    * the appropriate environment variable: */
   
   if (opt_pvfstab_set) {
      if ((setenv("PVFS2TAB_FILE", opt_pvfs2tab, 1)) < 0) {
         perror("setenv");
         goto die_jar_jar_die;
      }
   }
   
   /* this is how much of the file data is covered on each iteration of
    * the test.  used to help determine the seek offset on each
    * iteration */
   iter_jump = nprocs * opt_block;
      
   /* setup a buffer of data to write */
   if (!(tmp = malloc((size_t) opt_block + 256))) {
      perror("malloc");
      goto die_jar_jar_die;
   }
   buf = tmp + 128 - (((long)tmp) % 128);  /* align buffer */

   /* open the file for writing */
	if (opt_coll) {
		comm = MPI_COMM_WORLD;
	}
	else {
		comm = MPI_COMM_SELF;
	}
   err = MPI_File_open(comm, opt_file, 
							  MPI_MODE_CREATE | MPI_MODE_RDWR, MPI_INFO_NULL, &fh);
   if (err != MPI_SUCCESS) {
      handle_error(err, "MPI_File_open");
      goto die_jar_jar_die;
   } 
	
	nchars = (int) (opt_block/sizeof(char));
	if (!opt_rdonly) {

   /* now repeat the seek and write operations the number of times
    * specified on the command line */
   for (j=0; j < opt_iter; j++) {

      /* reading and writing to the same block is cheating, but sometimes
       * we want to measure cached performance of file servers */
      if (opt_single == 1)
			seek_position = 0;
      else
			/* seek to an appropriate position depending on the iteration 
			 * and rank of the current process */
			seek_position = (j*iter_jump)+(mynod*opt_block);

      MPI_File_seek(fh, seek_position, MPI_SEEK_SET);

      if (opt_correct) /* fill in buffer for iteration */ {
         for (i=0, v=mynod+j, check=buf; i<opt_block; i++, v++, check++) 
            *check = (char) v;
      }

      /* discover the starting time of the operation */
      MPI_Barrier(MPI_COMM_WORLD);
      stim = MPI_Wtime();

      /* write out the data */
		if (opt_coll) {
			err = MPI_File_write_all(fh, buf, nchars, MPI_CHAR, &status);
		}
		else {
			err = MPI_File_write(fh, buf, nchars, MPI_CHAR, &status);
		}
      if(err){
         fprintf(stderr, "node %d, write error: %s\n", mynod, 
         strerror(errno));
      }
      if (opt_sync) sync_err = MPI_File_sync(fh);
      if (sync_err) {
         fprintf(stderr, "node %d, sync error: %s\n", mynod, 
					  strerror(errno));
      }

      /* discover the ending time of the operation */
      etim = MPI_Wtime();

      write_tim += (etim - stim);
      
      /* we are done with this "write" iteration */
   }
	} /* ! opt_rdonly */

   err = MPI_File_close(&fh);
   if(err){
      fprintf(stderr, "node %d, close error after write\n", mynod);
   }
    
   /* wait for everyone to synchronize at this point */
   MPI_Barrier(MPI_COMM_WORLD);

   /* reopen the file to read the data back out */
   err = MPI_File_open(comm, opt_file, 
			   MPI_MODE_CREATE | MPI_MODE_RDWR, MPI_INFO_NULL, &fh);
   if (err < 0) {
      fprintf(stderr, "node %d, open error: %s\n", mynod, strerror(errno));
      goto die_jar_jar_die;
   }

	if (!opt_wronly) {
   /* we are going to repeat the read operation the number of iterations
    * specified */
   for (j=0; j < opt_iter; j++) {
      /* reading and writing to the same block is cheating, but sometimes
       * we want to measure cached performance of file servers */
      if (opt_single == 1) {
			seek_position = 0;
		}
      else {
			/* seek to an appropriate position depending on the iteration 
			 * and rank of the current process */
			seek_position = (j*iter_jump)+(mynod*opt_block);
		}

      MPI_File_seek(fh, seek_position, MPI_SEEK_SET);

      /* discover the start time */
      MPI_Barrier(MPI_COMM_WORLD);
      stim = MPI_Wtime();

      /* read in the file data */
		if (opt_coll) {
			err = MPI_File_read_all(fh, buf, nchars, MPI_CHAR, &status);
		}
		else {
			err = MPI_File_read(fh, buf, nchars, MPI_CHAR, &status);
		}
      myerrno = errno;

      /* discover the end time */
      etim = MPI_Wtime();
      read_tim += (etim - stim);

      if (err < 0) {
			fprintf(stderr, "node %d, read error, loc = %lld: %s\n",
				mynod, (long long) mynod*opt_block, strerror(myerrno));
		}

      /* if the user wanted to check correctness, compare the write
       * buffer to the read buffer */
      if (opt_correct) {
         int badct = 0;

         for (i=0, v=mynod+j, check=buf;
              i < opt_block && badct < 10;
              i++, v++, check++)
         {
            if (*check != (char) v) {
               my_correct = 0;
               if (badct < 10) {
                  badct++;
                  fprintf(stderr, "buf[%d] = %d, should be %d\n", 
                          i, *check, (char) v);
               }
            }
         }
         MPI_Allreduce(&my_correct, &correct, 1, MPI_INT, MPI_MIN,
                       MPI_COMM_WORLD);
         if (badct == 10) fprintf(stderr, "...\n");
      }

      /* we are done with this read iteration */
   }
	} /* !opt_wronly */

   /* close the file */
   err = MPI_File_close(&fh);
   if (err) {
      fprintf(stderr, "node %d, close error after write\n", mynod);
   }

   /* compute the read and write times */
   MPI_Allreduce(&read_tim, &max_read_tim, 1, MPI_DOUBLE, MPI_MAX,
      MPI_COMM_WORLD);
   MPI_Allreduce(&read_tim, &min_read_tim, 1, MPI_DOUBLE, MPI_MIN,
      MPI_COMM_WORLD);
   MPI_Allreduce(&read_tim, &sum_read_tim, 1, MPI_DOUBLE, MPI_SUM,
      MPI_COMM_WORLD);

   /* calculate our part of the summation used for variance */
   sq_read_tim = read_tim - (sum_read_tim / nprocs);
   sq_read_tim = sq_read_tim * sq_read_tim;
   MPI_Allreduce(&sq_read_tim, &sumsq_read_tim, 1, MPI_DOUBLE, 
                   MPI_SUM, MPI_COMM_WORLD);


   MPI_Allreduce(&write_tim, &max_write_tim, 1, MPI_DOUBLE, MPI_MAX,
      MPI_COMM_WORLD);
   MPI_Allreduce(&write_tim, &min_write_tim, 1, MPI_DOUBLE, MPI_MIN,
      MPI_COMM_WORLD);
   MPI_Allreduce(&write_tim, &sum_write_tim, 1, MPI_DOUBLE, MPI_SUM,
      MPI_COMM_WORLD);

   /* calculate our part of the summation used for variance */
   sq_write_tim = write_tim - (sum_write_tim / nprocs );
   sq_write_tim = sq_write_tim * sq_write_tim;
   MPI_Allreduce(&sq_write_tim, &sumsq_write_tim, 1, MPI_DOUBLE, 
                   MPI_SUM, MPI_COMM_WORLD);

   /* calculate the average from the sum */
   ave_read_tim  = sum_read_tim / nprocs; 
   ave_write_tim = sum_write_tim / nprocs; 

   /* and finally compute variance */
   if (nprocs > 1) {
		var_read_tim  = sumsq_read_tim / (nprocs-1);
		var_write_tim = sumsq_write_tim / (nprocs-1);
   }
   else {
		var_read_tim = 0;
		var_write_tim = 0;
   }
   
   /* print out the results on one node */
   if (mynod == 0) {
      read_bw = (opt_block*nprocs*opt_iter)/(max_read_tim*1.0e6);
      write_bw = (opt_block*nprocs*opt_iter)/(max_write_tim*1.0e6);
      
		printf("nr_procs = %d, nr_iter = %d, blk_sz = %lld, coll = %d\n",
				 nprocs, opt_iter, (long long) opt_block, opt_coll);
		
		printf("# total_size = %lld\n",
				 (long long) opt_block*nprocs*opt_iter);
		printf("# Write: min_t = %f, max_t = %f, mean_t = %f, var_t = %f\n", 
				 min_write_tim, max_write_tim, ave_write_tim, var_write_tim);
		printf("# Read:  min_t = %f, max_t = %f, mean_t = %f, var_t = %f\n", 
				 min_read_tim, max_read_tim, ave_read_tim, var_read_tim);
      
      printf("Write bandwidth = %f Mbytes/sec\n", write_bw);
      printf("Read bandwidth = %f Mbytes/sec\n", read_bw);
      
      if (opt_correct) {
         printf("Correctness test %s.\n", correct ? "passed" : "failed");
      }
   }


die_jar_jar_die:   

   free(tmp);
   MPI_Finalize();
   return(0);
}

static int parse_args(int argc, char **argv)
{
   int c;
   
   while ((c = getopt(argc, argv, "b:i:f:p:CcyShvrw")) != EOF) {
      switch (c) {
         case 'b': /* block size */
            opt_block = atoi(optarg);
            break;
         case 'i': /* iterations */
            opt_iter = atoi(optarg);
            break;
         case 'f': /* filename */
            strncpy(opt_file, optarg, 255);
            break;
         case 'p': /* pvfstab file */
            strncpy(opt_pvfs2tab, optarg, 255);
            opt_pvfstab_set = 1;
            break;
         case 'c': /* correctness */
            opt_correct = 1;
            break;
         case 'C': /* collective I/O */
            opt_coll = 1;
            break;
         case 'y': /* sYnc */
            opt_sync = 1;
            break;
         case 'S': /* Single region */
            opt_single = 1;
            break;
         case 'v': /* verbose */
            opt_verbose = 1;
            break;
			case 'r': /* read-only */
				opt_rdonly = 1;
				break;
			case 'w': /* write-only */
				opt_wronly = 1;
				break;
         case 'h':
            if (mynod == 0)
                usage();
            exit(0);
         case '?': /* unknown */
            if (mynod == 0)
                usage();
            exit(1);
         default:
            break;
      }
   }
   return(0);
}

static void usage(void)
{
    printf("Usage: mpi-io-test [<OPTIONS>...]\n");
    printf("\n<OPTIONS> is one of\n");
    printf(" -b       block size (in bytes) [default: 16777216]\n");
    printf(" -c       verify correctness of file data [default: off]\n");
    printf(" -C       perform operations Collectively [default: off]\n");
    printf(" -i       iterations [default: 1]\n");
    printf(" -f       filename [default: /foo/test.out]\n");
    printf(" -p       path to pvfs2tab file to use [default: notset]\n");
    printf(" -S       all process write to same Single region of file [default: off]\n");
	 printf(" -r       read-only.  do no writes.  file must already exist\n");
	 printf(" -w       write-only. do no reads.\n");
    printf(" -v       be more verbose\n");
    printf(" -y       sYnc the file after each write [default: off]\n");
    printf(" -h       print this help\n");
}

static void handle_error(int errcode, char *str)
{
    char msg[MPI_MAX_ERROR_STRING];
    int resultlen;

    MPI_Error_string(errcode, msg, &resultlen);
    fprintf(stderr, "%s: %s\n", str, msg);
    MPI_Abort(MPI_COMM_WORLD, 1);
}

/*
 * Local variables:
 *  c-indent-level: 3
 *  c-basic-offset: 3
 *  tab-width: 3
 *
 * vim: ts=3
 * End:
 */