diff options
author | Bill Wendling <wendling@ncsa.uiuc.edu> | 2001-12-19 22:10:15 (GMT) |
---|---|---|
committer | Bill Wendling <wendling@ncsa.uiuc.edu> | 2001-12-19 22:10:15 (GMT) |
commit | c9782c380b9f7b25dd13be7d3533b8be784b85b6 (patch) | |
tree | 86dd5aeb21073b4f1fc2678923d5a4f0eb6975e3 /perform/pio_perf.c | |
parent | 5beabb3b75cdf5c11dfdd1d305113d284ab64e0f (diff) | |
download | hdf5-c9782c380b9f7b25dd13be7d3533b8be784b85b6.zip hdf5-c9782c380b9f7b25dd13be7d3533b8be784b85b6.tar.gz hdf5-c9782c380b9f7b25dd13be7d3533b8be784b85b6.tar.bz2 |
[svn-r4746]
Purpose:
Feature Fix
Description:
Changed so the "pio_perf" module handles creating and destroying the
MPI Comm. Worked it so we get the minimum, maximum, and average times
over a set of iterations.
Solution:
Lots. Had to pull the MPI Comm code from the "pio_engine" module and
place it in the "pio_perf" module. Then worked on a way to have all
processes send their time output to process 0, who collects it and
gives back the min, max, and avg times for the iterations.
Platforms tested:
Linux. Doesn't work if you use more than 1 processor...*hrmph*
Diffstat (limited to 'perform/pio_perf.c')
-rw-r--r-- | perform/pio_perf.c | 388 |
1 files changed, 317 insertions, 71 deletions
diff --git a/perform/pio_perf.c b/perform/pio_perf.c index 8439f36..731df9b 100644 --- a/perform/pio_perf.c +++ b/perform/pio_perf.c @@ -50,6 +50,7 @@ */ /* system header files */ +#include <stdarg.h> #include <stdio.h> #include <stdlib.h> @@ -80,6 +81,12 @@ #define MIN_HDF5_BUF_SIZE (ONE_MB >> 1) #define MAX_HDF5_BUF_SIZE (ONE_GB / 2) +/* global variables */ +MPI_Comm pio_comm_g; /* Communicator to run the PIO */ +int pio_mpi_rank_g; /* MPI rank of pio_comm_g */ +int pio_mpi_nprocs_g; /* number of processes of pio_comm_g */ + + /* local variables */ static const char *progname = "pio_perf"; @@ -196,11 +203,23 @@ struct options { long min_xfer_size; /* minimum transfer buffer size */ }; +typedef struct _minmax { + double min; + double max; + double sum; + int num; +} minmax; + /* local functions */ static long parse_size_directive(const char *size); static struct options *parse_command_line(int argc, char *argv[]); static void run_test_loop(FILE *output, struct options *options); -static void run_test(FILE *output, iotype iot, parameters parms); +static int run_test(FILE *output, iotype iot, parameters parms); +static void get_minmax(minmax *mm); +static minmax accumulate_minmax_stuff(minmax *mm, int count); +static int create_comm_world(int num_procs); +static int destroy_comm_world(void); +static void output_report(FILE *output, const char *fmt, ...); static void print_indent(register FILE *output, register int indent); static void usage(const char *prog); @@ -218,17 +237,7 @@ main(int argc, char **argv) int world_size, ret; int exit_value = EXIT_SUCCESS; FILE *output = stdout; - struct options *opts; - - opts = parse_command_line(argc, argv); - - if (opts->output_file) { - if ((output = fopen(opts->output_file, "w")) == NULL) { - fprintf(stderr, "%s: cannot open output file\n", progname); - perror(opts->output_file); - goto onions; - } - } + struct options *opts = NULL; /* initialize MPI and get the maximum num of processors we started with */ MPI_Init(&argc, &argv); @@ -243,15 +252,30 @@ main(int argc, char **argv) fprintf(stderr, "invalid argument\n"); exit_value = EXIT_FAILURE; - goto cheese_and; + goto finish; + } + + pio_comm_g = MPI_COMM_WORLD; + + opts = parse_command_line(argc, argv); + + if (!opts) { + exit_value = EXIT_FAILURE; + goto finish; + } + + if (opts->output_file) { + if ((output = fopen(opts->output_file, "w")) == NULL) { + fprintf(stderr, "%s: cannot open output file\n", progname); + perror(opts->output_file); + goto finish; + } } run_test_loop(output, opts); -cheese_and: +finish: MPI_Finalize(); - -onions: free(opts); return exit_value; } @@ -301,24 +325,31 @@ run_test_loop(FILE *output, struct options *opts) parms.num_dsets = opts->num_dsets; parms.num_iters = opts->num_iters; - /* divide the maximum number of processors by 2 for each loop iter */ + /* multiply the maximum number of processors by 2 for each loop iter */ for (num_procs = opts->min_num_procs; num_procs <= opts->max_num_procs; num_procs <<= 1) { - register long j; + register long buf_size; parms.num_procs = num_procs; - fprintf(output, "Number of processors = %u\n", parms.num_procs); - for (j = opts->min_xfer_size; j <= opts->max_xfer_size; j <<= 1) { - parms.buf_size = j; + if (create_comm_world(parms.num_procs) != SUCCESS) { + /* do something harsh */ + } + + output_report(output, "Number of processors = %ld\n", parms.num_procs); + + /* multiply the xfer buffer size by 2 for each loop iteration */ + for (buf_size = opts->min_xfer_size; + buf_size <= opts->max_xfer_size; buf_size <<= 1) { + parms.buf_size = buf_size; parms.num_elmts = opts->file_size / (parms.num_dsets * sizeof(int)); print_indent(output, TAB_SPACE * 1); - fprintf(output, "Transfer Buffer Size: %u\n", j); + output_report(output, "Transfer Buffer Size: %ld\n", buf_size); print_indent(output, TAB_SPACE * 1); - fprintf(output, - " # of files: %u, # of dsets: %lu, Elements per dset: %lu\n", - parms.num_files, parms.num_dsets, parms.num_elmts); + output_report(output, + " # of files: %ld, # of dsets: %ld, # of elmts per dset: %ld\n", + parms.num_files, parms.num_dsets, parms.num_elmts); if (io_runs & PIO_RAW) run_test(output, RAW, parms); @@ -329,6 +360,10 @@ run_test_loop(FILE *output, struct options *opts) if (io_runs & PIO_HDF5) run_test(output, PHDF5, parms); } + + if (destroy_comm_world() != SUCCESS) { + /* do something harsh */ + } } } @@ -339,41 +374,248 @@ run_test_loop(FILE *output, struct options *opts) * Programmer: Bill Wendling, 18. December 2001 * Modifications: */ -static void +static int run_test(FILE *output, iotype iot, parameters parms) { - results res; + results res; + register int i, ret_value = SUCCESS; + int comm_size; + minmax total_mm; + minmax *write_mm_table; + minmax *read_mm_table; + minmax write_mm = {0.0, 0.0, 0.0, 0}; + minmax read_mm = {0.0, 0.0, 0.0, 0}; parms.io_type = iot; print_indent(output, TAB_SPACE * 2); - fprintf(output, "Type of IO = "); + output_report(output, "Type of IO = "); switch (iot) { case RAW: - fprintf(output, "Raw\n"); + output_report(output, "Raw\n"); break; case MPIO: - fprintf(output, "MPIO\n"); + output_report(output, "MPIO\n"); break; case PHDF5: - fprintf(output, "PHDF5\n"); + output_report(output, "PHDF5\n"); break; } + MPI_Comm_size(pio_comm_g, &comm_size); + + write_mm_table = malloc(parms.num_iters * sizeof(minmax)); + read_mm_table = malloc(parms.num_iters * sizeof(minmax)); + + for (i = 0; i < parms.num_iters; ++i) { + write_mm_table[i].min = 0.0; + write_mm_table[i].max = 0.0; + write_mm_table[i].sum = 0.0; + write_mm_table[i].num = 0; + + read_mm_table[i].min = 0.0; + read_mm_table[i].max = 0.0; + read_mm_table[i].sum = 0.0; + read_mm_table[i].num = 0; + } + /* call Albert's testing here */ - res = do_pio(parms); + for (i = 0; i < parms.num_iters; ++i) { + register int j; + double t; + + MPI_Barrier(pio_comm_g); + res = do_pio(parms); + + t = get_time(res.timers, HDF5_GROSS_WRITE_FIXED_DIMS); + MPI_Send((void *)&t, 1, MPI_DOUBLE, 0, 0, pio_comm_g); + + for (j = 0; j < comm_size; ++j) + get_minmax(&write_mm); + + write_mm_table[i] = write_mm; + + t = get_time(res.timers, HDF5_GROSS_READ_FIXED_DIMS); + MPI_Send((void *)&t, 1, MPI_DOUBLE, 0, 0, pio_comm_g); + + for (j = 0; j < comm_size; ++j) + get_minmax(&read_mm); + + read_mm_table[i] = read_mm; + } + + total_mm = accumulate_minmax_stuff(write_mm_table, parms.num_iters); + +printf("write metrics: min: %f, max: %f, avg: %f\n", total_mm.min, + total_mm.max, total_mm.sum / total_mm.num); + + total_mm = accumulate_minmax_stuff(read_mm_table, parms.num_iters); + +printf("read metrics: min: %f, max: %f, avg: %f\n", total_mm.min, + total_mm.max, total_mm.sum / total_mm.num); print_indent(output, TAB_SPACE * 3); - fprintf(output, "Write Results = %.2f MB/s\n", - MB_PER_SEC(parms.num_dsets * parms.num_elmts * sizeof(int), - get_time(res.timers, HDF5_WRITE_FIXED_DIMS))); + output_report(output, "Write Results = %.2f MB/s\n", + MB_PER_SEC(parms.num_dsets * parms.num_elmts * sizeof(int), + get_time(res.timers, HDF5_GROSS_WRITE_FIXED_DIMS))); print_indent(output, TAB_SPACE * 3); - fprintf(output, "Read Results = %.2f MB/s\n", - MB_PER_SEC(parms.num_dsets * parms.num_elmts * sizeof(int), - get_time(res.timers, HDF5_READ_FIXED_DIMS))); + output_report(output, "Read Results = %.2f MB/s\n", + MB_PER_SEC(parms.num_dsets * parms.num_elmts * sizeof(int), + get_time(res.timers, HDF5_GROSS_READ_FIXED_DIMS))); + free(write_mm_table); + free(read_mm_table); pio_time_destroy(res.timers); + return ret_value; +} + +static void +get_minmax(minmax *mm) +{ + int myrank; + + MPI_Comm_rank(pio_comm_g, &myrank); + + if (myrank == 0) { + MPI_Status status = {0}; + double t; + + MPI_Recv((void *)&t, 1, MPI_DOUBLE, MPI_ANY_SOURCE, + MPI_ANY_TAG, pio_comm_g, &status); + + ++mm->num; + mm->sum += t; + + if (t > mm->max) + mm->max = t; + + if (t < mm->min || mm->min <= 0.0) + mm->min = t; + } +} + +static minmax +accumulate_minmax_stuff(minmax *mm, int count) +{ + register int i; + minmax total_mm = mm[0]; + + for (i = 1; i < count; ++i) { + total_mm.sum += mm[i].sum; + total_mm.num += mm[i].num; + + if (mm[i].min < total_mm.min) + total_mm.min = mm[i].min; + + if (mm[i].max > total_mm.max) + total_mm.max = mm[i].max; + } + + return total_mm; +} + +/* + * Function: create_comm_world + * Purpose: Create an MPI Comm world and store it in pio_comm_g, which + * is a global variable. + * Return: SUCCESS on success. + * FAIL otherwise. + * Programmer: Bill Wendling, 19. December 2001 + * Modifications: + */ +static int +create_comm_world(int num_procs) +{ + /* MPI variables */ + int mrc, ret_value; /* return values */ + int color; /* for communicator creation */ + int myrank, nprocs; + + pio_comm_g = MPI_COMM_NULL; + + /* + * Create a sub communicator for this PIO run. Easier to use the first N + * processes. + */ + MPI_Comm_size(MPI_COMM_WORLD, &nprocs); + + if (num_procs > nprocs) { + fprintf(stderr, + "number of process(%d) must be <= number of processes in MPI_COMM_WORLD(%d)\n", + num_procs, nprocs); + goto error_done; + } + + MPI_Comm_rank(MPI_COMM_WORLD, &myrank); + color = (myrank < num_procs); + mrc = MPI_Comm_split(MPI_COMM_WORLD, color, myrank, &pio_comm_g); + + if (mrc != MPI_SUCCESS) { + fprintf(stderr, "MPI_Comm_split failed\n"); + goto error_done; + } + + if (!color) { + /* not involved in this run */ + mrc = destroy_comm_world(); + goto done; + } + + /* determine the MPI rank in the PIO communicator */ + MPI_Comm_size(pio_comm_g, &pio_mpi_nprocs_g); + MPI_Comm_rank(pio_comm_g, &pio_mpi_rank_g); + +done: + return ret_value; + +error_done: + destroy_comm_world(); + return FAIL; +} + +/* + * Function: destroy_comm_world + * Purpose: Destroy the created MPI Comm world which is stored in the + * pio_comm_g global variable. + * Return: SUCCESS on success. + * FAIL otherwise. + * Programmer: Bill Wendling, 19. December 2001 + * Modifications: + */ +static int +destroy_comm_world(void) +{ + int mrc = SUCCESS; /* return code */ + + /* release MPI resources */ + if (pio_comm_g != MPI_COMM_NULL) + mrc = (MPI_Comm_free(&pio_comm_g) == MPI_SUCCESS ? SUCCESS : FAIL); + + return mrc; +} + +/* + * Function: output_report + * Purpose: Print a line of the report. Only do so if I'm the 0 process. + * Return: Nothing + * Programmer: Bill Wendling, 19. December 2001 + * Modifications: + */ +static void +output_report(FILE *output, const char *fmt, ...) +{ + int myrank; + + MPI_Comm_rank(pio_comm_g, &myrank); + + if (myrank == 0) { + va_list ap; + + va_start(ap, fmt); + vfprintf(output, fmt, ap); + va_end(ap); + } } /* @@ -465,13 +707,11 @@ parse_command_line(int argc, char *argv[]) cl_opts->max_xfer_size = parse_size_directive(opt_arg); break; case 'h': - usage(progname); - exit(EXIT_SUCCESS); case '?': default: - /* there could be other command line options, such as MPI stuff - * that gets passed to our program, for some reason */ - break; + usage(progname); + free(cl_opts); + return NULL; } } @@ -536,34 +776,40 @@ parse_size_directive(const char *size) static void usage(const char *prog) { - fflush(stdout); - fprintf(stdout, "usage: %s [OPTIONS]\n", prog); - fprintf(stdout, " OPTIONS\n"); - fprintf(stdout, " -h, --help Print a usage message and exit\n"); - fprintf(stdout, " -d N, --num-dsets=N Number of datasets per file [default:1]\n"); - fprintf(stdout, " -f S, --file-size=S Size of a single file [default: 64M]\n"); - fprintf(stdout, " -F N, --num-files=N Number of files [default: 1]\n"); - fprintf(stdout, " -H, --hdf5 Run HDF5 performance test\n"); - fprintf(stdout, " -i, --num-iterations Number of iterations to perform [default: 1]\n"); - fprintf(stdout, " -m, --mpiio Run MPI/IO performance test\n"); - fprintf(stdout, " -o F, --output=F Output raw data into file F [default: none]\n"); - fprintf(stdout, " -P N, --max-num-processes=N Maximum number of processes to use [default: 1]\n"); - fprintf(stdout, " -p N, --min-num-processes=N Minimum number of processes to use [default: 1]\n"); - fprintf(stdout, " -r, --raw Run raw (UNIX) performance test\n"); - fprintf(stdout, " -X S, --max-xfer-size=S Maximum transfer buffer size [default: 1M]\n"); - fprintf(stdout, " -x S, --min-xfer-size=S Minimum transfer buffer size [default: 1K]\n"); - fprintf(stdout, "\n"); - fprintf(stdout, " F - is a filename.\n"); - fprintf(stdout, " N - is an integer >=0.\n"); - fprintf(stdout, " S - is a size specifier, an integer >=0 followed by a size indicator:\n"); - fprintf(stdout, "\n"); - fprintf(stdout, " K - Kilobyte\n"); - fprintf(stdout, " M - Megabyte\n"); - fprintf(stdout, " G - Gigabyte\n"); - fprintf(stdout, "\n"); - fprintf(stdout, " Example: 37M = 37 Megabytes\n"); - fprintf(stdout, "\n"); - fflush(stdout); + int myrank; + + MPI_Comm_rank(pio_comm_g, &myrank); + + if (myrank == 0) { + fflush(stdout); + fprintf(stdout, "usage: %s [OPTIONS]\n", prog); + fprintf(stdout, " OPTIONS\n"); + fprintf(stdout, " -h, --help Print a usage message and exit\n"); + fprintf(stdout, " -d N, --num-dsets=N Number of datasets per file [default:1]\n"); + fprintf(stdout, " -f S, --file-size=S Size of a single file [default: 64M]\n"); + fprintf(stdout, " -F N, --num-files=N Number of files [default: 1]\n"); + fprintf(stdout, " -H, --hdf5 Run HDF5 performance test\n"); + fprintf(stdout, " -i, --num-iterations Number of iterations to perform [default: 1]\n"); + fprintf(stdout, " -m, --mpiio Run MPI/IO performance test\n"); + fprintf(stdout, " -o F, --output=F Output raw data into file F [default: none]\n"); + fprintf(stdout, " -P N, --max-num-processes=N Maximum number of processes to use [default: 1]\n"); + fprintf(stdout, " -p N, --min-num-processes=N Minimum number of processes to use [default: 1]\n"); + fprintf(stdout, " -r, --raw Run raw (UNIX) performance test\n"); + fprintf(stdout, " -X S, --max-xfer-size=S Maximum transfer buffer size [default: 1M]\n"); + fprintf(stdout, " -x S, --min-xfer-size=S Minimum transfer buffer size [default: 1K]\n"); + fprintf(stdout, "\n"); + fprintf(stdout, " F - is a filename.\n"); + fprintf(stdout, " N - is an integer >=0.\n"); + fprintf(stdout, " S - is a size specifier, an integer >=0 followed by a size indicator:\n"); + fprintf(stdout, "\n"); + fprintf(stdout, " K - Kilobyte\n"); + fprintf(stdout, " M - Megabyte\n"); + fprintf(stdout, " G - Gigabyte\n"); + fprintf(stdout, "\n"); + fprintf(stdout, " Example: 37M = 37 Megabytes\n"); + fprintf(stdout, "\n"); + fflush(stdout); + } } #else /* H5_HAVE_PARALLEL */ |