summaryrefslogtreecommitdiffstats
path: root/perform/pio_perf.c
blob: 4fda6a17728620ad77868938d645474557208346 (plain)
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
/*
 * Copyright (C) 2001
 *     National Center for Supercomputing Applications
 *     All rights reserved.
 *
 */

/*
 * Parallel HDF5 Performance Testing Code
 * --------------------------------------
 *
 * Portable code to test performance on the different platforms we support.
 * This is what the report should look like:
 *
 *  nprocs = Max#Procs
 *      IO Type = Raw
 *          # Files = 1, # of dsets = 1000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *          # Files = 1, # of dsets = 3000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *
 *          . . .
 *
 *      IO Type = MPIO
 *          # Files = 1, # of dsets = 1000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *          # Files = 1, # of dsets = 3000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *
 *          . . .
 *
 *      IO Type = PHDF5
 *          # Files = 1, # of dsets = 1000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *          # Files = 1, # of dsets = 3000, Elements per dset = 37000
 *              Write Results = x MB/s
 *              Read Results = x MB/s
 *
 *          . . .
 *
 *  nprocs = Max#Procs / 2
 *
 *      . . .
 *
 */

/* system header files */
#include <stdio.h>
#include <stdlib.h>

#include "hdf5.h"

#ifdef H5_HAVE_PARALLEL

/* library header files */
#include <mpi.h>

/* our header files */
#include "h5tools_utils.h"
#include "pio_perf.h"

/* useful macros */
#define TAB_SPACE           4

#define ONE_KB              1024
#define ONE_MB              (ONE_KB * ONE_KB)
#define ONE_GB              (ONE_MB * ONE_KB)

#define MB_PER_SEC(bytes,t) (((bytes) / ONE_MB) / t)

#define MIN_HDF5_BUF_SIZE   (ONE_MB >> 1)
#define MAX_HDF5_BUF_SIZE   (ONE_GB / 2)

/* local variables */
static const char  *progname = "pio_perf";

/*
 * Command-line options: The user can specify short or long-named
 * parameters. The long-named ones can be partially spelled. When
 * adding more, make sure that they don't clash with each other.
 */
#if 1
static const char *s_opts = "ho:";
#else
static const char *s_opts = "hbo:";
#endif  /* 1 */
static struct long_options l_opts[] = {
    { "help", no_arg, 'h' },
    { "hel", no_arg, 'h' },
    { "he", no_arg, 'h' },
#if 0
    /* a siting of the elusive binary option */
    { "binary", no_arg, 'b' },
    { "binar", no_arg, 'b' },
    { "bina", no_arg, 'b' },
    { "bin", no_arg, 'b' },
    { "bi", no_arg, 'b' },
#endif  /* 0 */
    { "max-size", require_arg, 'm' },
    { "max-siz", require_arg, 'm' },
    { "max-si", require_arg, 'm' },
    { "max-s", require_arg, 'm' },
    { "max", require_arg, 'm' },
    { "ma", require_arg, 'm' },
    { "output", require_arg, 'o' },
    { "outpu", require_arg, 'o' },
    { "outp", require_arg, 'o' },
    { "out", require_arg, 'o' },
    { "ou", require_arg, 'o' },
    { NULL, 0, '\0' }
};

struct options {
    const char *output_file;    /* file to print report to              */
    long max_size;              /* maximum size of file in gigabytes    */
};

/* local functions */
static struct options *parse_command_line(int argc, char *argv[]);
static void run_test_loop(FILE *output, int max_num_procs, long max_size);
static void print_indent(register FILE *output, register int indent);
static void usage(const char *prog);

/*
 * Function:    main
 * Purpose:     Start things up. Initialize MPI and then call the test looping
 *              function.
 * Return:      EXIT_SUCCESS or EXIT_FAILURE
 * Programmer:  Bill Wendling, 30. October 2001
 * Modifications:
 */
int
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;
        }
    }

    /* initialize MPI and get the maximum num of processors we started with */
    MPI_Init(&argc, &argv);
    ret = MPI_Comm_size(MPI_COMM_WORLD, &world_size);

    if (ret != MPI_SUCCESS) {
        fprintf(stderr, "%s: MPI_Comm_size call failed\n", progname);

        if (ret == MPI_ERR_COMM)
            fprintf(stderr, "invalid MPI communicator\n");
        else
            fprintf(stderr, "invalid argument\n");

        exit_value = EXIT_FAILURE;
        goto cheese_and;
    }

    run_test_loop(output, world_size, opts->max_size);

cheese_and:
    MPI_Finalize();

onions:
    free(opts);
    return exit_value;
}

/*
 * Function:    run_test_loop
 * Purpose:     Run the I/O tests. Write the results to OUTPUT.
 *
 *            - The slowest changing part of the test is the number of
 *              processors to use. For each loop iteration, we divide that
 *              number by 2 and rerun the test.
 *
 *            - The second slowest is what type of IO to perform. We have
 *              three choices: RAW, MPI-IO, and PHDF5.
 *
 *            - Then we change the size of the buffer. This information is
 *              inferred from the number of datasets to create and the number
 *              of integers to put into each dataset. The backend code figures
 *              this out.
 *
 * Return:      Nothing
 * Programmer:  Bill Wendling, 30. October 2001
 * Modifications:
 */
static void
run_test_loop(FILE *output, int max_num_procs, long max_size)
{
    parameters parms;

    /* num_files stays ``1'' for now but may change later */
    parms.num_files = 1;
    parms.num_iters = 1;

    /* divide the maximum number of processors by 2 for each loop iter */
    for (; max_num_procs > 0; max_num_procs /= 2) {
        register iotype i;

        parms.max_num_procs = max_num_procs;
        fprintf(output, "Number of processors = %u\n", parms.max_num_procs);

        for (i = RAW; i <= PHDF5; ++i) {
            register unsigned long j;

            parms.io_type = i;
            print_indent(output, TAB_SPACE * 1);
            fprintf(output, "Type of IO = ");

            if (i == RAW)
                fprintf(output, "Raw\n");
            else if (i == MPIO)
                fprintf(output, "MPIO\n");
            else
                fprintf(output, "PHDF5\n");

            for (j = MIN_HDF5_BUF_SIZE; j <= MAX_HDF5_BUF_SIZE; j <<= 1) {
                results res;

                parms.buf_size = j;
                parms.num_dsets = ONE_GB / j;
                parms.num_elmts = (max_size * j) / sizeof(int);

                print_indent(output, TAB_SPACE * 2);
                fprintf(output,
                        "# of files: %u, # of dsets: %lu, Elements per dset: %lu\n",
                        parms.num_files, parms.num_dsets, parms.num_elmts);

                /* call Albert's testing here */
                res = do_pio(parms); 

                print_indent(output, TAB_SPACE * 3);
                fprintf(output, "Write Results = %f MB/s\n",
                        MB_PER_SEC(parms.num_dsets * parms.num_elmts * sizeof(int),
                                   get_time(res.timers, HDF5_WRITE_FIXED_DIMS)));

                pio_time_destroy(res.timers);
            }
        }
    }
}

/*
 * Function:    print_indent
 * Purpose:     Print spaces to indent a new line of text for pretty printing
 *              things.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 29. October 2001
 * Modifications:
 */
static void
print_indent(register FILE *output, register int indent)
{
    for (; indent > 0; --indent)
        fputc(' ', output);
}

/*
 * Function:    usage
 * Purpose:     Print a usage message and then exit.
 * Return:      Nothing
 * Programmer:  Bill Wendling, 31. October 2001
 * Modifications:
 */
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, "     -m #, --max-size=#   Maximum size of file in gigabytes [default: 2]\n");
    fprintf(stdout, "     -o F, --output=F     Output raw data into file F\n");
    fprintf(stdout, "\n");
    fprintf(stdout, "  F - is a filename.\n");
    fprintf(stdout, "\n");
}

/*
 * Function:    parse_command_line
 * Purpose:     Parse the command line options and return a STRUCT OPTIONS
 *              structure which will need to be freed by the calling function.
 * Return:      Pointer to an OPTIONS structure
 * Programmer:  Bill Wendling, 31. October 2001
 * Modifications:
 */
static struct options *
parse_command_line(int argc, char *argv[])
{
    int opt;
    struct options *cl_opts;

    cl_opts = (struct options *)calloc(1, sizeof(struct options));
    cl_opts->max_size = 2;

    while ((opt = get_option(argc, (const char **)argv, s_opts, l_opts)) != EOF) {
        switch ((char)opt) {
#if 0
        case 'b':
            /* the future "binary" option */
            break;
#endif  /* 0 */
        case 'm':
            cl_opts->max_size = atol(opt_arg);
            break;
        case 'o':
            cl_opts->output_file = 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;
        }
    }

    return cl_opts;
}

#else /* H5_HAVE_PARALLEL */

/*
 * Function:    main
 * Purpose:     Dummy main() function for if HDF5 was configured without
 *              parallel stuff.
 * Return:      EXIT_SUCCESS
 * Programmer:  Bill Wendling, 14. November 2001
 * Modifications:
 */
int
main(void)
{
    printf("No parallel IO performance because parallel is not configured\n");
    return EXIT_SUCCESS;
}

#endif /* !H5_HAVE_PARALLEL */