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
|
/*
Copyright 2009, UCAR/Unidata
See COPYRIGHT file for copying and redistribution conditions.
This program tests netcdf-4 parallel I/O. These tests are based on the
needs of the NASA GMAO model, and are based on some test code from
Dennis Nadeau.
$Id: tst_nc4perf.c,v 1.4 2009/08/19 15:58:57 ed Exp $
*/
#include "nc_tests.h"
#include "err_macros.h"
#define FILENAME "tst_nc4perf.nc"
#define NDIMS1 2
#define NDIMS2 4
#define DIMSIZE1 40
#define DIMSIZE2 61
#define DIMSIZE3 3
/*#define DIMSIZE1 540
#define DIMSIZE2 361
#define DIMSIZE3 72*/
#define TIMELEN 4
#define NUMVARS 10
#define NUM_TRIES 2
#define MEGABYTE 1048576
/* This function creates a file with 10 2D variables, no unlimited
* dimension. */
int test_pio_2d(size_t cache_size, int facc_type, int access_flag, MPI_Comm comm,
MPI_Info info, int mpi_size, int mpi_rank,
size_t *chunk_size)
{
double starttime, endtime, write_time = 0, bandwidth = 0;
int ncid;
int dimids[NDIMS1];
size_t start[NDIMS1], count[NDIMS1];
float *data;
char file_name[NC_MAX_NAME + 1];
char var_name1[NUMVARS][NC_MAX_NAME + 1] = {"GWa", "JAd", "TJe", "JMa", "JMo",
"JQA", "AJa", "MVB", "WHH", "JTy"};
int varid1[NUMVARS];
size_t nelems_in;
float preemption_in;
int j, i, t;
/* Create some data. */
if (!(data = malloc(sizeof(float) * DIMSIZE2 * DIMSIZE1 / mpi_size)))
return -2;
for (j = 0; j < DIMSIZE2; j++)
for (i = 0; i < DIMSIZE1 / mpi_size; i++)
data[j * DIMSIZE1 / mpi_size + i] = (float)mpi_rank * (j + 1);
/* Get the file name. */
sprintf(file_name, "%s/%s", TEMP_LARGE, FILENAME);
/* Set the cache size. */
if (nc_get_chunk_cache(NULL, &nelems_in, &preemption_in)) ERR;
if (nc_set_chunk_cache(cache_size, nelems_in, preemption_in)) ERR;
for (t = 0; t < NUM_TRIES; t++)
{
/* Create a netcdf-4 file, opened for parallel I/O. */
if (nc_create_par(file_name, facc_type|NC_NETCDF4, comm,
info, &ncid)) ERR;
/* Create two dimensions. */
if (nc_def_dim(ncid, "d1", DIMSIZE2, &dimids[0])) ERR;
if (nc_def_dim(ncid, "d2", DIMSIZE1, &dimids[1])) ERR;
/* Create our variables. */
for (i = 0; i < NUMVARS; i++)
{
if (nc_def_var(ncid, var_name1[i], NC_INT, NDIMS1,
dimids, &varid1[i])) ERR;
if (chunk_size[0])
if (nc_def_var_chunking(ncid, varid1[i], 0, chunk_size)) ERR;
}
if (nc_enddef(ncid)) ERR;
/* Set up slab for this process. */
start[0] = 0;
start[1] = mpi_rank * DIMSIZE1/mpi_size;
count[0] = DIMSIZE2;
count[1] = DIMSIZE1 / mpi_size;
/* start parallel netcdf4 */
for (i = 0; i < NUMVARS; i++)
if (nc_var_par_access(ncid, varid1[i], access_flag)) ERR;
starttime = MPI_Wtime();
/* Write two dimensional float data */
for (i = 0; i < NUMVARS; i++)
if (nc_put_vara_float(ncid, varid1[i], start, count, data)) ERR;
/* Close the netcdf file. */
if (nc_close(ncid)) ERR;
endtime = MPI_Wtime();
if (!mpi_rank)
{
bandwidth += ((sizeof(float) * DIMSIZE1 * DIMSIZE2 * NUMVARS) /
((endtime - starttime) * 1024 * 1024)) / NUM_TRIES;
write_time += (endtime - starttime) / NUM_TRIES;
}
}
free(data);
if (!mpi_rank)
{
char chunk_string[NC_MAX_NAME + 1] = "";
/* What was our chunking? */
if (chunk_size[0])
sprintf(chunk_string, "%dx%d ", (int)chunk_size[0], (int)chunk_size[1]);
else
strcat(chunk_string, "contiguous");
/* Print the results. */
printf("%d\t\t%s\t%s\t%d\t\t%dx%d\t\t%s\t%f\t\t%f\t\t\t%d\n", mpi_size,
(facc_type == NC_MPIIO ? "MPI-IO " : "MPI-POSIX"),
(access_flag == NC_INDEPENDENT ? "independent" : "collective"),
(int)cache_size/MEGABYTE, DIMSIZE1, DIMSIZE2, chunk_string,
write_time, bandwidth, NUM_TRIES);
}
/* Delete this file. */
remove(file_name);
return 0;
}
/* Both read and write will be tested */
/* Case 2: create four dimensional integer data,
one dimension is unlimited. */
int test_pio_4d(size_t cache_size, int facc_type, int access_flag, MPI_Comm comm,
MPI_Info info, int mpi_size, int mpi_rank, size_t *chunk_size)
{
int ncid, dimuids[NDIMS2], varid2[NUMVARS];
size_t ustart[NDIMS2], ucount[NDIMS2];
float *udata, *tempudata;
char file_name[NC_MAX_NAME + 1];
char var_name2[NUMVARS][NC_MAX_NAME + 1] = {"JKP", "ZTa", "MFi", "FPi", "JBu",
"ALi", "AJo", "USG", "RBH", "JAG"};
double starttime, endtime, write_time = 0, bandwidth = 0;
size_t nelems_in;
float preemption_in;
int k, j, i, t;
udata = malloc(DIMSIZE3 * DIMSIZE2 * DIMSIZE1 / mpi_size * sizeof(int));
/* Create phony data. */
tempudata = udata;
for(k = 0; k < DIMSIZE3; k++)
for(j = 0; j < DIMSIZE2; j++)
for(i = 0; i < DIMSIZE1 / mpi_size; i++)
{
*tempudata = (float)(1 + mpi_rank) * 2 * (j + 1) * (k + 1);
tempudata++;
}
/* Get the file name. */
sprintf(file_name, "%s/%s", TEMP_LARGE, FILENAME);
/* Set the cache size. */
if (nc_get_chunk_cache(NULL, &nelems_in, &preemption_in)) ERR;
if (nc_set_chunk_cache(cache_size, nelems_in, preemption_in)) ERR;
for (t = 0; t < NUM_TRIES; t++)
{
/* Create a netcdf-4 file. */
if (nc_create_par(file_name, facc_type|NC_NETCDF4, comm, info,
&ncid)) ERR;
/* Create four dimensions. */
if (nc_def_dim(ncid, "ud1", TIMELEN, dimuids)) ERR;
if (nc_def_dim(ncid, "ud2", DIMSIZE3, &dimuids[1])) ERR;
if (nc_def_dim(ncid, "ud3", DIMSIZE2, &dimuids[2])) ERR;
if (nc_def_dim(ncid, "ud4", DIMSIZE1, &dimuids[3])) ERR;
/* Create 10 variables. */
for (i = 0; i < NUMVARS; i++)
if (nc_def_var(ncid, var_name2[i], NC_INT, NDIMS2,
dimuids, &varid2[i])) ERR;
if (nc_enddef(ncid)) ERR;
/* Set up selection parameters */
ustart[0] = 0;
ustart[1] = 0;
ustart[2] = 0;
ustart[3] = DIMSIZE1 * mpi_rank / mpi_size;
ucount[0] = 1;
ucount[1] = DIMSIZE3;
ucount[2] = DIMSIZE2;
ucount[3] = DIMSIZE1 / mpi_size;
/* Access parallel */
for (i = 0; i < NUMVARS; i++)
if (nc_var_par_access(ncid, varid2[i], access_flag)) ERR;
starttime = MPI_Wtime();
/* Write slabs of phony data. */
for(ustart[0] = 0; ustart[0] < TIMELEN; ustart[0]++)
for (i = 0; i < NUMVARS; i++)
if (nc_put_vara_float(ncid, varid2[i], ustart, ucount, udata)) ERR;
/* Close the netcdf file. */
if (nc_close(ncid)) ERR;
endtime = MPI_Wtime();
if (!mpi_rank)
{
write_time += (endtime - starttime) / NUM_TRIES;
bandwidth += (sizeof(float) * TIMELEN * DIMSIZE1 * DIMSIZE2 * DIMSIZE3 * NUMVARS) /
((endtime - starttime) * 1024 * 1024 * NUM_TRIES);
}
}
free(udata);
if (!mpi_rank)
{
char chunk_string[NC_MAX_NAME + 1] = "";
/* What was our chunking? */
if (chunk_size[0])
sprintf(chunk_string, "%dx%dx%dx%d", (int)chunk_size[0], (int)chunk_size[1],
(int)chunk_size[2], (int)chunk_size[3]);
else
strcat(chunk_string, "contiguous");
/* Print our results. */
printf("%d\t\t%s\t%s\t%d\t\t%dx%dx%dx%d\t%s\t%f\t\t%f\t\t\t%d\n", mpi_size,
(facc_type == NC_MPIIO ? "MPI-IO " : "MPI-POSIX"),
(access_flag == NC_INDEPENDENT ? "independent" : "collective"),
(int)cache_size / MEGABYTE, TIMELEN, DIMSIZE3, DIMSIZE2, DIMSIZE1, chunk_string, write_time,
bandwidth, NUM_TRIES);
}
/* Delete this file. */
remove(file_name);
return 0;
}
/* Note: When the MPI-POSIX VFD is not compiled in to HDF5, the NC_MPIPOSIX
* flag will be aliased to the NC_MPIIO flag within the library, and
* therefore this test will exercise the aliasing, with the MPI-IO VFD,
* under that configuration. -QAK
*/
#define NUM_MODES 2
#define NUM_FACC 2
#define NUM_CHUNK_COMBOS_2D 3
#define NUM_CHUNK_COMBOS_4D 4
#define NUM_CACHE_SIZES 3
int main(int argc, char **argv)
{
MPI_Comm comm = MPI_COMM_WORLD;
MPI_Info info = MPI_INFO_NULL;
int mpi_size, mpi_rank;
int mpi_mode[NUM_MODES] = {NC_MPIIO, NC_MPIPOSIX};
int facc_type[NUM_FACC] = {NC_INDEPENDENT, NC_COLLECTIVE};
size_t chunk_size_2d[NUM_CHUNK_COMBOS_2D][NDIMS1] = {{0, 0},
{DIMSIZE2, DIMSIZE1},
{DIMSIZE2/2 + 1, DIMSIZE1 / 2}};
size_t chunk_size_4d[NUM_CHUNK_COMBOS_4D][NDIMS2] = {{0, 0, 0, 0},
{1, DIMSIZE3, DIMSIZE2, DIMSIZE1},
{TIMELEN / 2, DIMSIZE3 / 2 + 1, DIMSIZE2 / 2 + 1, DIMSIZE1 / 2},
{TIMELEN, DIMSIZE3, DIMSIZE2, DIMSIZE1}};
size_t cache_size[NUM_CACHE_SIZES] = {MEGABYTE, 32 * MEGABYTE, 64 * MEGABYTE};
int m, f, c, i;
/* Initialize MPI. */
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
/* Check for invalid number of processors. */
if ((float)DIMSIZE1 / mpi_size != (int)(DIMSIZE1 / mpi_size))
{
printf("%d divided by number of processors must be a whole number!\n",
DIMSIZE1);
return -1;
}
if (!mpi_rank)
{
printf("*** Testing parallel IO for NASA...\n");
printf("num_proc\tMPI mode\taccess\t\tcache (MB)\tgrid size\tchunks\tavg. write time(s)\t"
"avg. write bandwidth(MB/s)\tnum_tries\n");
}
for (i = 0; i < NUM_CACHE_SIZES; i++)
for (m = 0; m < NUM_MODES; m++)
for (f = 0; f < NUM_FACC; f++)
for (c = 0; c < NUM_CHUNK_COMBOS_2D; c++)
if (test_pio_2d(cache_size[i], mpi_mode[m], facc_type[f], comm,
info, mpi_size, mpi_rank, chunk_size_2d[c])) ERR;
for (i = 0; i < NUM_CACHE_SIZES; i++)
for (m = 0; m < NUM_MODES; m++)
for (f = 0; f < NUM_FACC; f++)
for (c = 0; c < NUM_CHUNK_COMBOS_4D; c++)
if (test_pio_4d(cache_size[i], mpi_mode[m], facc_type[f], comm,
info, mpi_size, mpi_rank, chunk_size_4d[c])) ERR;
if (!mpi_rank)
SUMMARIZE_ERR;
MPI_Finalize();
if (!mpi_rank)
FINAL_RESULTS;
return 0;
}
|
