codes-darshan-io-wrkld.c 59.3 KB
Newer Older
1
2
3
4
5
/*
 * Copyright (C) 2013 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
 */
6
#include <assert.h>
7
8

#include "codes/codes-workload.h"
9
#include "codes/quickhash.h"
10
#include "codes-workload-method.h"
11

12
#include "darshan-logutils.h"
13

14
15
16
#define DEF_INTER_IO_DELAY_PCT 0.2
#define DEF_INTER_CYC_DELAY_PCT 0.4

17
18
19
20
#define DARSHAN_NEGLIGIBLE_DELAY .001

#define RANK_HASH_TABLE_SIZE 397

21
22
23
24
25
#define IO_IS_IN_SIZE_BIN_RANGE(size, bin_ndx, bin_min_sizes)                       \
        ((bin_ndx == 9) ?                                                           \
        (size >= bin_min_sizes[bin_ndx]) :                                          \
        ((size >= bin_min_sizes[bin_ndx]) && (size < bin_min_sizes[bin_ndx + 1])))

26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
/* structure for storing a darshan workload operation (a codes op with 2 timestamps) */
struct darshan_io_op
{
    struct codes_workload_op codes_op;
    double start_time;
    double end_time;
};

/* I/O context structure managed by each rank in the darshan workload */
struct rank_io_context
{
    int64_t my_rank;
    double last_op_time;
    void *io_op_dat;
    struct qhash_head hash_link;
};

/* Darshan workload generator's implementation of the CODES workload API */
44
45
static int darshan_io_workload_load(const char *params, int rank);
static void darshan_io_workload_get_next(int rank, struct codes_workload_op *op);
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
static void *darshan_io_workload_get_info(int rank);
static int darshan_rank_hash_compare(void *key, struct qhash_head *link);

/* Darshan I/O op data structure access (insert, remove) abstraction */
static void *darshan_init_io_op_dat(void);
static void darshan_insert_next_io_op(void *io_op_dat, struct darshan_io_op *io_op);
static void darshan_remove_next_io_op(void *io_op_dat, struct darshan_io_op *io_op,
                                      double last_op_time);
static void darshan_finalize_io_op_dat(void *io_op_dat);
static int darshan_io_op_compare(const void *p1, const void *p2);

/* Helper functions for implementing the (complex, nonfactored) Darshan workload generator */
static void generate_psx_ind_file_events(struct darshan_file *file,
                                         struct rank_io_context *io_context);
static void generate_psx_coll_file_events(struct darshan_file *file,
                                          struct rank_io_context *io_context,
                                          int64_t nprocs, int64_t aggregator_cnt);
63
static double generate_psx_open_event(struct darshan_file *file, int create_flag,
64
65
                                      double meta_op_time, double cur_time,
                                      struct rank_io_context *io_context);
66
static double generate_psx_close_event(struct darshan_file *file, double meta_op_time,
67
68
69
                                       double cur_time, struct rank_io_context *io_context);
static double generate_barrier_event(struct darshan_file *file, int64_t root, double cur_time,
                                     struct rank_io_context *io_context);
70
71
static double generate_psx_ind_io_events(struct darshan_file *file, int64_t io_ops_this_cycle,
                                         int64_t open_ndx, double inter_io_delay, 
72
73
74
75
76
77
78
                                         double meta_op_time, double cur_time,
                                         struct rank_io_context *io_context);
static double generate_psx_coll_io_events(struct darshan_file *file, int64_t ind_io_ops_this_cycle,
                                          int64_t coll_io_ops_this_cycle, int64_t nprocs,
                                          int64_t aggregator_cnt, int64_t open_ndx,
                                          double inter_io_delay, double meta_op_time, double cur_time,
                                          struct rank_io_context *io_context);
79
80
81
82
83
84
85
static void determine_io_params(struct darshan_file *file, int write_flag, int coll_flag,
                                int64_t io_cycles, size_t *io_sz, off_t *io_off);
static void calc_io_delays(struct darshan_file *file, int64_t num_opens, int64_t num_io_ops,
                           double delay_per_cycle, double *first_io_delay, double *close_delay,
                           double *inter_open_delay, double *inter_io_delay);
static void file_sanity_check(struct darshan_file *file, struct darshan_job *job);

86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
/* helper functions for implementing the darshan workload generator */
static void generate_psx_ind_file_events(struct darshan_file *file);
static double generate_psx_open_event(struct darshan_file *file, int create_flag,
                                      double meta_op_time, double cur_time);
static double generate_psx_close_event(struct darshan_file *file, double meta_op_time,
                                       double cur_time);
static double generate_barrier_event(struct darshan_file *file, int64_t root, double cur_time);
static double generate_psx_ind_io_events(struct darshan_file *file, int64_t io_ops_this_cycle,
                                         int64_t open_ndx, double inter_io_delay, 
                                         double meta_op_time, double cur_time);
static void determine_io_params(struct darshan_file *file, int write_flag, int coll_flag,
                                int64_t io_cycles, size_t *io_sz, off_t *io_off);
static void calc_io_delays(struct darshan_file *file, int64_t num_opens, int64_t num_io_ops,
                           double delay_per_cycle, double *first_io_delay, double *close_delay,
                           double *inter_open_delay, double *inter_io_delay);
static void file_sanity_check(struct darshan_file *file, struct darshan_job *job);

103
104
105
106
107
108
109
110
/* workload method name and function pointers for the CODES workload API */
struct codes_workload_method darshan_io_workload_method =
{
    .method_name = "darshan_io_workload",
    .codes_workload_load = darshan_io_workload_load,
    .codes_workload_get_next = darshan_io_workload_get_next,
};

111
/* info about this darshan workload group needed by bgp model */
112
113
/* TODO: is this needed for darshan workloads? */
/* TODO: does this need to be stored in the rank context to support multiple workloads? */
114

115
/* hash table to store per-rank workload contexts */
116
static struct qhash_table *rank_tbl = NULL;
117

118
119
120
/* load the workload generator for this rank, given input params */
static int darshan_io_workload_load(const char *params, int rank)
{
121
122
    darshan_params *d_params = (darshan_params *)params;
    darshan_fd logfile_fd;
123
124
    struct darshan_job job;
    struct darshan_file next_file;
125
    struct rank_io_context *my_ctx;
126
    int ret;
127

128
    if (!d_params)
129
130
        return -1;

131
132
    /* (re)seed the random number generator */
    srand(time(NULL));
133

134
    /* open the darshan log to begin reading in file i/o info */
135
136
    logfile_fd = darshan_log_open(d_params->log_file_path, "r");
    if (logfile_fd < 0)
137
        return -1;
138

139
140
141
142
143
144
145
146
    /* get the per-job stats from the log */
    ret = darshan_log_getjob(logfile_fd, &job);
    if (ret < 0)
    {
        darshan_log_close(logfile_fd);
        return -1;
    }

147
148
149
150
151
152
153
154
155
156
157
    /* allocate the i/o context needed by this rank */
    my_ctx = malloc(sizeof(struct rank_io_context));
    if (!my_ctx)
    {
        darshan_log_close(logfile_fd);
        return -1;
    }
    my_ctx->my_rank = (int64_t)rank;
    my_ctx->last_op_time = 0.0;
    my_ctx->io_op_dat = darshan_init_io_op_dat();

158
159
160
161
162
163
164
165
166
167
    /* loop over all files contained in the log file */
    while ((ret = darshan_log_getfile(logfile_fd, &job, &next_file)) > 0)
    {
        /* generate all i/o events contained in this independent file */
        if (next_file.rank == rank)
        {
            /* make sure the file i/o counters are valid */
            file_sanity_check(&next_file, &job);

            /* generate i/o events and store them in this rank's workload context */
168
            generate_psx_ind_file_events(&next_file, my_ctx);
169
170
171
172
173
174
        }
        /* generate all i/o events involving this rank in this collective file */
        else if (next_file.rank == -1)
        {
            /* make sure the file i/o counters are valid */
            file_sanity_check(&next_file, &job);
175
176
177

            /* generate collective i/o events and store them in the rank context */
            generate_psx_coll_file_events(&next_file, my_ctx, job.nprocs, d_params->aggregator_cnt);
178
179
        }
    }
180

181
182
183
    if (ret < 0)
        return -1;

184
    darshan_log_close(logfile_fd);
185

186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
    /* finalize the rank's i/o context so i/o ops may be retrieved later (in order) */
    darshan_finalize_io_op_dat(my_ctx->io_op_dat);

    /* initialize the hash table of rank contexts, if it has not been initialized */
    if (!rank_tbl)
    {
        rank_tbl = qhash_init(darshan_rank_hash_compare, quickhash_64bit_hash, RANK_HASH_TABLE_SIZE);
        if (!rank_tbl)
            return -1;
    }

    /* add this rank context to the hash table */
    qhash_add(rank_tbl, &(my_ctx->my_rank), &(my_ctx->hash_link));

    /* fill out the info required for this workload group */
    if (darshan_workload_info.group_id == -1)
    {
        darshan_workload_info.group_id = 1;
        darshan_workload_info.min_rank = 0;
        darshan_workload_info.max_rank = job.nprocs - 1;
        darshan_workload_info.num_lrank = job.nprocs;
    }

209
210
211
212
213
214
    return 0;
}

/* pull the next event (independent or collective) for this rank from its event context */
static void darshan_io_workload_get_next(int rank, struct codes_workload_op *op)
{
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
    int64_t my_rank = (int64_t)rank;
    struct qhash_head *hash_link = NULL;
    struct rank_io_context *tmp = NULL;
    struct darshan_io_op next_io_op;

    /* find i/o context for this rank in the rank hash table */
    hash_link = qhash_search(rank_tbl, &my_rank);

    /* terminate the workload if there is no valid rank context */
    if (!hash_link)
    {
        op->op_type = CODES_WK_END;
        return;
    }

    /* get access to the rank's io_context data */
    tmp = qhash_entry(hash_link, struct rank_io_context, hash_link);
    assert(tmp->my_rank == my_rank);

    /* get the next darshan i/o op out of this rank's context */
    darshan_remove_next_io_op(tmp->io_op_dat, &next_io_op, tmp->last_op_time);

    /* free the rank's i/o context if this is the last i/o op */
    if (next_io_op.codes_op.op_type == CODES_WK_END)
    {
        qhash_del(hash_link);
        free(tmp);
    }
    else
    {
        /* else, set the last op time to be the end of the returned op */
        tmp->last_op_time = next_io_op.end_time;
    }

    /* return the codes op contained in the darshan i/o op */
    *op = next_io_op.codes_op;
251
252
253
254

    return;
}

255
256
257
/* return the workload info needed by the bgp model */
/* TODO: do we really need this? */
static void *darshan_io_workload_get_info(int rank)
258
{
259
    return &(darshan_workload_info);
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
/* comparison function for comparing two hash keys (used for storing multiple io contexts) */
static int darshan_rank_hash_compare(
    void *key, struct qhash_head *link)
{
    int64_t *in_rank = (int64_t *)key;
    struct rank_io_context *tmp;

    tmp = qhash_entry(link, struct rank_io_context, hash_link);
    if (tmp->my_rank == *in_rank)
        return 1;

    return 0;
}

/*****************************************/
/*                                       */
/*   Darshan I/O op storage abstraction  */
/*                                       */
/*****************************************/

#define DARSHAN_IO_OP_INC_CNT 100000

/* dynamically allocated array data structure for storing darshan i/o events */
struct darshan_io_dat_array
{
    struct darshan_io_op *op_array;
    int64_t op_arr_ndx;
    int64_t op_arr_cnt;
};

/* initialize the dynamic array data structure */
static void *darshan_init_io_op_dat()
{
    struct darshan_io_dat_array *tmp;

    /* initialize the array data structure */
    tmp = malloc(sizeof(struct darshan_io_dat_array));
    assert(tmp);
    tmp->op_array = malloc(DARSHAN_IO_OP_INC_CNT * sizeof(struct darshan_io_op));
    assert(tmp->op_array);
    tmp->op_arr_ndx = 0;
    tmp->op_arr_cnt = DARSHAN_IO_OP_INC_CNT;

    /* return the array info for this rank's i/o context */
    return (void *)tmp;
}

/* store the i/o event in this rank's i/o context */
static void darshan_insert_next_io_op(
    void *io_op_dat, struct darshan_io_op *io_op)
{
    struct darshan_io_dat_array *array = (struct darshan_io_dat_array *)io_op_dat;
    struct darshan_io_op *tmp;

    /* realloc array if it is already full */
    if (array->op_arr_ndx == array->op_arr_cnt)
    {
        tmp = malloc((array->op_arr_cnt + DARSHAN_IO_OP_INC_CNT) * sizeof(struct darshan_io_op));
        assert(tmp);
        memcpy(tmp, array->op_array, array->op_arr_cnt * sizeof(struct darshan_io_op));
        free(array->op_array);
        array->op_array = tmp;
        array->op_arr_cnt += DARSHAN_IO_OP_INC_CNT;
    }

    /* add the darshan i/o op to the array */
    array->op_array[array->op_arr_ndx++] = *io_op;

    return;
}

/* pull the next i/o event out of this rank's i/o context */
static void darshan_remove_next_io_op(
    void *io_op_dat, struct darshan_io_op *io_op, double last_op_time)
{
    struct darshan_io_dat_array *array = (struct darshan_io_dat_array *)io_op_dat;

    /* if the array has been scanned completely already */
    if (array->op_arr_ndx == array->op_arr_cnt)
    {
        /* no more events just end the workload */
        io_op->codes_op.op_type = CODES_WK_END;

        /* free data structures */
        free(array->op_array);
        free(array);
    }
    else
    {
        struct darshan_io_op *tmp = &(array->op_array[array->op_arr_ndx]);

        if ((tmp->start_time - last_op_time) < DARSHAN_NEGLIGIBLE_DELAY)
        {
            /* there is no delay, just return the next op in the array */
            *io_op = *tmp;
            array->op_arr_ndx++;
        }
        else
        {
            /* there is a nonnegligible delay, so generate and return a delay event */
            io_op->codes_op.op_type = CODES_WK_DELAY;
            io_op->codes_op.u.delay.seconds = tmp->start_time - last_op_time;
            io_op->start_time = last_op_time;
            io_op->end_time = tmp->start_time;
        }
    }
}

/* sort the dynamic array in order of i/o op start time */
static void darshan_finalize_io_op_dat(
    void *io_op_dat)
{
    struct darshan_io_dat_array *array = (struct darshan_io_dat_array *)io_op_dat;

    /* sort this rank's i/o op list */
    qsort(array->op_array, array->op_arr_ndx, sizeof(struct darshan_io_op), darshan_io_op_compare);
    array->op_arr_cnt = array->op_arr_ndx;
    array->op_arr_ndx = 0;

    return;
}

/* comparison function for sorting darshan_io_ops in order of start timestamps */
static int darshan_io_op_compare(
    const void *p1, const void *p2)
{
    struct darshan_io_op *a = (struct darshan_io_op *)p1;
    struct darshan_io_op *b = (struct darshan_io_op *)p2;

    if (a->start_time < b->start_time)
        return -1;
    else if (a->start_time > b->start_time)
        return 1;
    else
        return 0;
}

/*****************************************/
/*                                       */
/* Darshan workload generation functions */
/*                                       */
/*****************************************/
404
405
406

/* generate events for an independently opened file, and store these events */
static void generate_psx_ind_file_events(
407
    struct darshan_file *file, struct rank_io_context *io_context)
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
{
    int64_t io_ops_this_cycle;
    double cur_time = file->fcounters[CP_F_OPEN_TIMESTAMP];
    double delay_per_open;
    double first_io_delay = 0.0;
    double close_delay = 0.0;
    double inter_open_delay = 0.0;
    double inter_io_delay = 0.0;
    double meta_op_time;
    int create_flag;
    int64_t i;

    /* if the file was never really opened, just return because we have no timing info */
    if (file->counters[CP_POSIX_OPENS] == 0)
        return;

    /* determine delay available per open-io-close cycle */
    delay_per_open = (file->fcounters[CP_F_CLOSE_TIMESTAMP] - file->fcounters[CP_F_OPEN_TIMESTAMP] -
                     file->fcounters[CP_F_POSIX_READ_TIME] - file->fcounters[CP_F_POSIX_WRITE_TIME] -
                     file->fcounters[CP_F_POSIX_META_TIME]) / file->counters[CP_POSIX_OPENS];

    /* calculate synthetic delay values */
    calc_io_delays(file, file->counters[CP_POSIX_OPENS],
                   file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES],
                   delay_per_open, &first_io_delay, &close_delay,
                   &inter_open_delay, &inter_io_delay);

    /* calculate average meta op time (for i/o and opens/closes) */
    /* TODO: this needs to be updated when we add in stat, seek, etc. */
    meta_op_time = file->fcounters[CP_F_POSIX_META_TIME] / ((2 * file->counters[CP_POSIX_OPENS]) +
                   file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES]);

440
441
442
443
444
    /* set the create flag if the file was written to */
    if (file->counters[CP_BYTES_WRITTEN])
    {
        create_flag = 1;
    }
445
446
447
448
449
450

    /* generate open/io/close events for all cycles */
    /* TODO: add stats */
    for (i = 0; file->counters[CP_POSIX_OPENS]; i++, file->counters[CP_POSIX_OPENS]--)
    {
        /* generate an open event */
451
        cur_time = generate_psx_open_event(file, create_flag, meta_op_time, cur_time, io_context);
452
453
454
455
456
457
458
459
460
461
462
        create_flag = 0;

        /* account for potential delay from first open to first io */
        cur_time += first_io_delay;

        io_ops_this_cycle = ceil((double)(file->counters[CP_POSIX_READS] +
                                 file->counters[CP_POSIX_WRITES]) /
                                 file->counters[CP_POSIX_OPENS]);

        /* perform the calculated number of i/o operations for this file open */
        cur_time = generate_psx_ind_io_events(file, io_ops_this_cycle, i, inter_io_delay,
463
                                              meta_op_time, cur_time, io_context);
464
465
466
467
468

        /* account for potential delay from last io to close */
        cur_time += close_delay;

        /* generate a close for the open event at the start of the loop */
469
        cur_time = generate_psx_close_event(file, meta_op_time, cur_time, io_context);
470
471
472
473
474
475
476

        /* account for potential interopen delay if more than one open */
        if (file->counters[CP_POSIX_OPENS] > 1)
        {
            cur_time += inter_open_delay;
        }
    }
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750

    return;
}

/* generate events for the i/o ops stored in a collectively opened file for this rank */
void generate_psx_coll_file_events(
    struct darshan_file *file, struct rank_io_context *io_context,
    int64_t nprocs, int64_t in_agg_cnt)
{
    int64_t total_io_ops = file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES];
    int64_t extra_opens = 0;
    int64_t extra_open_stride = 0;
    int64_t extra_io_ops = 0;
    int64_t ind_open_stride = 0;
    int64_t total_ind_io_ops;
    int64_t total_coll_io_ops;
    int64_t ind_io_ops_this_cycle;
    int64_t coll_io_ops_this_cycle;
    int64_t rank_cnt;
    int64_t aggregator_cnt;
    int create_flag = 0;
    double cur_time = file->fcounters[CP_F_OPEN_TIMESTAMP];
    double delay_per_cycle;
    double first_io_delay = 0.0;
    double close_delay = 0.0;
    double inter_cycle_delay = 0.0;
    double inter_io_delay = 0.0;
    double meta_op_time;
    int64_t i, j;

    /* the collective file was never opened (i.e., just stat-ed), so return */
    if (!(file->counters[CP_POSIX_OPENS]))
        return;

    /*  in this case, posix opens are less than mpi opens...
     *  this is probably a mpi deferred open -- assume app will not use this, currently.
     */
    assert(file->counters[CP_POSIX_OPENS] >= nprocs);

    /* determine delay information */
    delay_per_cycle = (file->fcounters[CP_F_CLOSE_TIMESTAMP] -
                      file->fcounters[CP_F_OPEN_TIMESTAMP] -
                      (file->fcounters[CP_F_POSIX_READ_TIME] / nprocs) -
                      (file->fcounters[CP_F_POSIX_WRITE_TIME] / nprocs) -
                      (file->fcounters[CP_F_POSIX_META_TIME] / nprocs)) /
                      (file->counters[CP_POSIX_OPENS] / nprocs);

    /* calculate average meta op time (for i/o and opens/closes) */
    meta_op_time = file->fcounters[CP_F_POSIX_META_TIME] / ((2 * file->counters[CP_POSIX_OPENS]) +
                   file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES]);

    /* TODO TIMING */ 
    calc_io_delays(file, (file->counters[CP_POSIX_OPENS] / nprocs),
                   round((double)total_io_ops / nprocs), delay_per_cycle,
                   &first_io_delay, &close_delay, &inter_cycle_delay, &inter_io_delay);

    if (file->counters[CP_COLL_OPENS] && file->counters[CP_INDEP_OPENS])
    {
        if (((file->counters[CP_COLL_OPENS] / nprocs) % file->counters[CP_INDEP_OPENS]) == 0)
        {        
            ind_open_stride = (file->counters[CP_COLL_OPENS] / nprocs) /
                              file->counters[CP_INDEP_OPENS];
        }
        else
        {
            ind_open_stride = 1;
        }
    }

    if (file->counters[CP_COLL_OPENS] || file->counters[CP_INDEP_OPENS])
    {
        extra_opens = file->counters[CP_POSIX_OPENS] - file->counters[CP_COLL_OPENS] -
                      file->counters[CP_INDEP_OPENS];
        assert(extra_opens <= ((file->counters[CP_COLL_OPENS] / nprocs) +
               file->counters[CP_INDEP_OPENS]));
        if (extra_opens)
        {
            if ((((file->counters[CP_COLL_OPENS] / nprocs) + file->counters[CP_INDEP_OPENS]) %
                extra_opens) == 0)
            {
                extra_open_stride = ((file->counters[CP_COLL_OPENS] / nprocs) +
                                    file->counters[CP_INDEP_OPENS]) / extra_opens;
            }
            else
            {
                extra_open_stride = 1;
            }
            extra_io_ops = 0;
        }

        total_ind_io_ops = file->counters[CP_INDEP_READS] + file->counters[CP_INDEP_WRITES];
        total_coll_io_ops = (file->counters[CP_COLL_READS] + file->counters[CP_COLL_WRITES]) / nprocs;
    }
    else 
    {
        extra_opens = file->counters[CP_POSIX_OPENS] % nprocs;
        if (extra_opens && (((file->counters[CP_POSIX_OPENS] / nprocs) % extra_opens) == 0))
        {
            extra_open_stride = (file->counters[CP_POSIX_OPENS] / nprocs) / extra_opens;
            extra_io_ops = total_io_ops % nprocs;
        }
        else
        {
            extra_opens = 0;
        }

        total_ind_io_ops = total_io_ops - extra_io_ops;
        total_coll_io_ops = 0;
    }

    /* it is rare to overwrite existing files, so set the create flag */
    if (file->counters[CP_BYTES_WRITTEN])
    {
        create_flag = 1;
    }

    /* generate all events for this collectively opened file */
    for (i = 0; file->counters[CP_POSIX_OPENS]; i++)
    {
        if (file->counters[CP_COLL_OPENS])
            aggregator_cnt = in_agg_cnt;
        else
            aggregator_cnt = nprocs;

        /* assign any determined 'extra' opens to rank 0 at the beginning of the cycle */
        if (extra_opens && !(i % extra_open_stride))
        {
            assert(create_flag);

            /* generate the open/close events for creating the collective file */
            file->rank = 0;
            cur_time = generate_psx_open_event(file, create_flag, meta_op_time, cur_time, io_context);

            ind_io_ops_this_cycle = ceil((double)extra_io_ops / extra_opens);

            cur_time = generate_psx_coll_io_events(file, ind_io_ops_this_cycle, 0, nprocs,
                                                   aggregator_cnt, i, 0.0, meta_op_time,
                                                   cur_time, io_context);
            extra_io_ops -= ind_io_ops_this_cycle;

            cur_time = generate_psx_close_event(file, meta_op_time, cur_time, io_context);
            create_flag = 0;
            file->rank = -1;
            file->counters[CP_POSIX_OPENS]--;
            extra_opens--;
        }

        if (file->counters[CP_POSIX_OPENS] >= nprocs)
        {
            if (file->counters[CP_COLL_OPENS])
            {
                if (file->counters[CP_INDEP_OPENS] && !(i % ind_open_stride))
                {
                    rank_cnt = ceil((double)file->counters[CP_INDEP_OPENS] /
                                    (file->counters[CP_COLL_OPENS] / nprocs));
                    for (j = 0; j < rank_cnt; j++)
                    {
                        file->rank = j;
                        if (j != (rank_cnt - 1))
                            generate_psx_open_event(file, 0, meta_op_time, cur_time, io_context);
                        else
                            cur_time = generate_psx_open_event(file, 0, meta_op_time,
                                                               cur_time, io_context);
                    }

                    ind_io_ops_this_cycle = ceil(((double)total_ind_io_ops /
                                            file->counters[CP_INDEP_OPENS]) * rank_cnt);

                    cur_time = generate_psx_coll_io_events(file, ind_io_ops_this_cycle, 0,
                                                           nprocs, aggregator_cnt, i, 0.0,
                                                           meta_op_time, cur_time, io_context);
                    total_ind_io_ops -= ind_io_ops_this_cycle;

                    for (j = 0; j < rank_cnt; j++)
                    {
                        file->rank = j;
                        if (j != (rank_cnt - 1))
                            generate_psx_close_event(file, meta_op_time, cur_time, io_context);
                        else
                            cur_time = generate_psx_close_event(file, meta_op_time,
                                                                cur_time, io_context);
                    }
                    file->rank = -1;
                    file->counters[CP_INDEP_OPENS] -= rank_cnt;
                    file->counters[CP_POSIX_OPENS] -= rank_cnt;
                }

                cur_time = generate_barrier_event(file, 0, cur_time, io_context);

                coll_io_ops_this_cycle = ceil((double)total_coll_io_ops /
                                         (file->counters[CP_COLL_OPENS] / nprocs));
                if (file->counters[CP_INDEP_OPENS])
                    ind_io_ops_this_cycle = 0;
                else
                    ind_io_ops_this_cycle = ceil((double)total_ind_io_ops /
                                            (file->counters[CP_POSIX_OPENS] / nprocs));
            }
            else
            {
                coll_io_ops_this_cycle = 0;
                ind_io_ops_this_cycle = ceil((double)total_ind_io_ops /
                                        (file->counters[CP_POSIX_OPENS] / nprocs));
            }

            /* perform an open across all ranks (rank == -1) */
            cur_time = generate_psx_open_event(file, create_flag, meta_op_time,
                                               cur_time, io_context);
            create_flag = 0;

            /* account for potential delay between the open and first i/o */
            cur_time += first_io_delay;

            /* generate the io events */
            cur_time = generate_psx_coll_io_events(file, ind_io_ops_this_cycle,
                                                   coll_io_ops_this_cycle, nprocs, aggregator_cnt,
                                                   i, inter_io_delay, meta_op_time,
                                                   cur_time, io_context);
            total_ind_io_ops -= ind_io_ops_this_cycle;
            total_coll_io_ops -= coll_io_ops_this_cycle;

            /* account for potential delay between last i/o operation and file close */
            cur_time += close_delay;
            
            /* generate the corresponding close event for the open at the start of the loop */
            cur_time = generate_psx_close_event(file, meta_op_time, cur_time, io_context);

            file->counters[CP_POSIX_OPENS] -= nprocs;
            if (file->counters[CP_COLL_OPENS])
                file->counters[CP_COLL_OPENS] -= nprocs;

            /* account for any delay between open-close cycles */
            if (file->counters[CP_POSIX_OPENS])
                cur_time += inter_cycle_delay;
        }
        else
        {
            /* open the file across participating ranks */
            rank_cnt = file->counters[CP_POSIX_OPENS];
            for (j = 0; j < rank_cnt; j++)
            {
                file->rank = j;
                if (j != (rank_cnt - 1))
                    generate_psx_open_event(file, 0, meta_op_time, cur_time, io_context);
                else
                    cur_time = generate_psx_open_event(file, 0, meta_op_time, cur_time, io_context);
            }

            /* account for potential delay between the open and first i/o */
            cur_time += first_io_delay;

            ind_io_ops_this_cycle = ceil((double)total_ind_io_ops /
                                         (file->counters[CP_POSIX_OPENS] / nprocs));

            cur_time = generate_psx_coll_io_events(file, ind_io_ops_this_cycle, 0, nprocs,
                                                   rank_cnt, i, inter_io_delay, meta_op_time,
                                                   cur_time, io_context);
            total_ind_io_ops -= ind_io_ops_this_cycle;

            /* account for potential delay between last i/o operation and file close */
            cur_time += close_delay;

            /* close the file across participating ranks */
            for (j = 0; j < rank_cnt; j++)
            {
                file->rank = j;
                if (j != (rank_cnt - 1))
                    generate_psx_close_event(file, meta_op_time, cur_time, io_context);
                else
                    cur_time = generate_psx_close_event(file, meta_op_time, cur_time, io_context);
            }
            file->rank = -1;
            file->counters[CP_POSIX_OPENS] -= rank_cnt;
        }
    }
751
752
753
754
755
756

    return;
}

/* fill in an open event structure and store it with the rank context */
static double generate_psx_open_event(
757
758
    struct darshan_file *file, int create_flag, double meta_op_time,
    double cur_time, struct rank_io_context *io_context)
759
{
760
761
762
763
764
765
766
    struct darshan_io_op next_io_op = 
    {
        .codes_op.op_type = CODES_WK_OPEN,
        .codes_op.u.open.file_id = file->hash,
        .codes_op.u.open.create_flag = create_flag,
        .start_time = cur_time
    };
767
768
769

    /* set the end time of the event based on time spent in POSIX meta operations */
    cur_time += meta_op_time;
770
    next_io_op.end_time = cur_time;
771

772
773
774
    /* store the open event (if this rank performed it) */
    if ((file->rank == io_context->my_rank) || (file->rank == -1))
        darshan_insert_next_io_op(io_context->io_op_dat, &next_io_op);
775
776
777
778
779
780

    return cur_time;
}

/* fill in a close event structure and store it with the rank context */
static double generate_psx_close_event(
781
782
    struct darshan_file *file, double meta_op_time, double cur_time,
    struct rank_io_context *io_context)
783
{
784
785
786
787
788
789
    struct darshan_io_op next_io_op =
    {
        .codes_op.op_type = CODES_WK_CLOSE,
        .codes_op.u.close.file_id = file->hash,
        .start_time = cur_time
    };
790
791
792

    /* set the end time of the event based on time spent in POSIX meta operations */
    cur_time += meta_op_time;
793
    next_io_op.end_time = cur_time;
794

795
796
797
    /* store the close event (if this rank performed it) */
    if ((file->rank == io_context->my_rank) || (file->rank == -1))
        darshan_insert_next_io_op(io_context->io_op_dat, &next_io_op);
798
799
800
801
802
803

    return cur_time;
}

/* fill in a barrier event structure and store it with the rank context */
static double generate_barrier_event(
804
    struct darshan_file *file, int64_t root, double cur_time, struct rank_io_context *io_context)
805
{
806
807
808
809
810
811
812
    struct darshan_io_op next_io_op =
    {
        .codes_op.op_type = CODES_WK_BARRIER, 
        .codes_op.u.barrier.count = -1, /* all processes */
        .codes_op.u.barrier.root = root,
        .start_time = cur_time
    };
813
814

    cur_time += .000001; /* small synthetic delay representing time to barrier */
815
    next_io_op.end_time = cur_time;
816

817
818
819
    /* store the barrier event */
    if ((file->rank == -1) || (file->rank == io_context->my_rank))
        darshan_insert_next_io_op(io_context->io_op_dat, &next_io_op);
820
821
822
823
824
825
826

    return cur_time;
}

/* generate all i/o events for one independent file open and store them with the rank context */
static double generate_psx_ind_io_events(
    struct darshan_file *file, int64_t io_ops_this_cycle, int64_t open_ndx,
827
    double inter_io_delay, double meta_op_time, double cur_time, struct rank_io_context *io_context)
828
829
830
831
832
833
834
835
836
{
    static int rw = -1; /* rw = 1 for write, 0 for read, -1 for uninitialized */
    static int64_t io_ops_this_rw;
    static double rd_bw = 0.0, wr_bw = 0.0;
    int64_t psx_rw_ops_remaining = file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES];
    double io_op_time;
    size_t io_sz;
    off_t io_off;
    int64_t i;
837
    struct darshan_io_op next_io_op;
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877

    /* if there are no i/o ops, just return immediately */
    if (!io_ops_this_cycle)
        return cur_time;

    /* initialze static variables when a new file is opened */
    if (rw == -1)
    {
        /* initialize rw to be the first i/o operation found in the log */
        if (file->fcounters[CP_F_WRITE_START_TIMESTAMP] == 0.0)
            rw = 0;
        else if (file->fcounters[CP_F_READ_START_TIMESTAMP] == 0.0)
            rw = 1;
        else
            rw = (file->fcounters[CP_F_READ_START_TIMESTAMP] <
                  file->fcounters[CP_F_WRITE_START_TIMESTAMP]) ? 0 : 1;

        /* determine how many io ops to do before next rw switch */
        if (!rw)
            io_ops_this_rw = file->counters[CP_POSIX_READS] /
                             ((file->counters[CP_RW_SWITCHES] / 2) + 1);
        else
            io_ops_this_rw = file->counters[CP_POSIX_WRITES] /
                             ((file->counters[CP_RW_SWITCHES] / 2) + 1);

        /* initialize the rd and wr bandwidth values using total io size and time */
        if (file->fcounters[CP_F_POSIX_READ_TIME])
            rd_bw = file->counters[CP_BYTES_READ] / file->fcounters[CP_F_POSIX_READ_TIME];
        if (file->fcounters[CP_F_POSIX_WRITE_TIME])
            wr_bw = file->counters[CP_BYTES_WRITTEN] / file->fcounters[CP_F_POSIX_WRITE_TIME];
    }

    /* loop to generate all reads/writes for this open/close sequence */
    for (i = 0; i < io_ops_this_cycle; i++)
    {
        /* calculate what value to use for i/o size and offset */
        determine_io_params(file, rw, 0, file->counters[CP_POSIX_OPENS], &io_sz, &io_off);
        if (!rw)
        {
            /* generate a read event */
878
879
880
881
882
            next_io_op.codes_op.op_type = CODES_WK_READ;
            next_io_op.codes_op.u.read.file_id = file->hash;
            next_io_op.codes_op.u.read.size = io_sz;
            next_io_op.codes_op.u.read.offset = io_off;
            next_io_op.start_time = cur_time;
883
884
885
886
887

            /* set the end time based on observed bandwidth and io size */
            if (rd_bw == 0.0)
                io_op_time = 0.0;
            else
888
                io_op_time = (io_sz / rd_bw);
889
890
891

            /* update time, accounting for metadata time */
            cur_time += (io_op_time + meta_op_time);
892
            next_io_op.end_time = cur_time;
893
894
895
896
897
            file->counters[CP_POSIX_READS]--;
        }
        else
        {
            /* generate a write event */
898
899
900
901
902
            next_io_op.codes_op.op_type = CODES_WK_WRITE;
            next_io_op.codes_op.u.write.file_id = file->hash;
            next_io_op.codes_op.u.write.size = io_sz;
            next_io_op.codes_op.u.write.offset = io_off;
            next_io_op.start_time = cur_time;
903
904
905
906
907

            /* set the end time based on observed bandwidth and io size */
            if (wr_bw == 0.0)
                io_op_time = 0.0;
            else
908
                io_op_time = (io_sz / wr_bw);
909
910
911

            /* update time, accounting for metadata time */
            cur_time += (io_op_time + meta_op_time);
912
            next_io_op.end_time = cur_time;
913
914
915
916
917
918
919
            file->counters[CP_POSIX_WRITES]--;
        }
        psx_rw_ops_remaining--;
        io_ops_this_rw--;
        assert(file->counters[CP_POSIX_READS] >= 0);
        assert(file->counters[CP_POSIX_WRITES] >= 0);

920
921
        /* store the i/o event */
        darshan_insert_next_io_op(io_context->io_op_dat, &next_io_op);
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941

        /* determine whether to toggle between reads and writes */
        if (!io_ops_this_rw && psx_rw_ops_remaining)
        {
            /* toggle the read/write flag */
            rw ^= 1;
            file->counters[CP_RW_SWITCHES]--;

            /* determine how many io ops to do before next rw switch */
            if (!rw)
                io_ops_this_rw = file->counters[CP_POSIX_READS] /
                                 ((file->counters[CP_RW_SWITCHES] / 2) + 1);
            else
                io_ops_this_rw = file->counters[CP_POSIX_WRITES] /
                                 ((file->counters[CP_RW_SWITCHES] / 2) + 1);
        }

        if (i != (io_ops_this_cycle - 1))
        {
            /* update current time to account for possible delay between i/o operations */
942
            cur_time += inter_io_delay;
943
944
945
946
947
948
949
950
951
952
953
954
        }
    }

    /* reset the static rw flag if this is the last open-close cycle for this file */
    if (file->counters[CP_POSIX_OPENS] == 1)
    {
        rw = -1;
    }

    return cur_time;
}

955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
static double generate_psx_coll_io_events(
    struct darshan_file *file, int64_t ind_io_ops_this_cycle, int64_t coll_io_ops_this_cycle,
    int64_t nprocs, int64_t aggregator_cnt, int64_t open_ndx, double inter_io_delay,
    double meta_op_time, double cur_time, struct rank_io_context *io_context)
{
    static int rw = -1; /* rw = 1 for write, 0 for read, -1 for uninitialized */
    static int64_t io_ops_this_rw;
    static double rd_bw = 0.0, wr_bw = 0.0;
    int64_t psx_rw_ops_remaining = file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES];
    int64_t total_io_ops_this_cycle = ind_io_ops_this_cycle + coll_io_ops_this_cycle;
    int64_t total_coll_io_ops =
            (file->counters[CP_COLL_READS] + file->counters[CP_COLL_WRITES]) / nprocs;
    int64_t tmp_rank;
    int64_t next_ind_io_rank = 0;
    int64_t io_cnt;
    int64_t ranks_per_aggregator = nprocs / aggregator_cnt;
    int64_t ind_ops_so_far = 0;
    double ind_coll_switch;
    double io_op_time;
    double max_cur_time = 0.0;
    int ind_coll;
    size_t io_sz;
    off_t io_off;
    int64_t i, j;
    struct darshan_io_op next_io_op;

    if (!total_io_ops_this_cycle)
        return cur_time;

    /* initialze static variables when a new file is opened */
    if (rw == -1)
    {
        /* initialize rw to be the first i/o operation found in the log */
        if (file->fcounters[CP_F_WRITE_START_TIMESTAMP] == 0.0)
            rw = 0;
        else if (file->fcounters[CP_F_READ_START_TIMESTAMP] == 0.0)
            rw = 1;
        else
            rw = (file->fcounters[CP_F_READ_START_TIMESTAMP] <
                  file->fcounters[CP_F_WRITE_START_TIMESTAMP]) ? 0 : 1;

        /* determine how many io ops to do before next rw switch */
        if (!rw)
        {
            if (file->counters[CP_COLL_OPENS])
                io_ops_this_rw =
                    ((file->counters[CP_COLL_READS] / nprocs) + file->counters[CP_INDEP_READS]) /
                    ((file->counters[CP_RW_SWITCHES] / (2 * aggregator_cnt)) + 1);
            else
                io_ops_this_rw = file->counters[CP_POSIX_READS] /
                                 ((file->counters[CP_RW_SWITCHES] / (2 * aggregator_cnt)) + 1);
        }
        else
        {
            if (file->counters[CP_COLL_OPENS])
                io_ops_this_rw =
                    ((file->counters[CP_COLL_WRITES] / nprocs) + file->counters[CP_INDEP_WRITES]) /
                    ((file->counters[CP_RW_SWITCHES] / (2 * aggregator_cnt)) + 1);
            else
                io_ops_this_rw = file->counters[CP_POSIX_WRITES] /
                                 ((file->counters[CP_RW_SWITCHES] / (2 * aggregator_cnt)) + 1);
        }

        /* initialize the rd and wr bandwidth values using total io size and time */
        if (file->fcounters[CP_F_POSIX_READ_TIME])
            rd_bw = file->counters[CP_BYTES_READ] / file->fcounters[CP_F_POSIX_READ_TIME];
        if (file->fcounters[CP_F_POSIX_WRITE_TIME])
            wr_bw = file->counters[CP_BYTES_WRITTEN] / file->fcounters[CP_F_POSIX_WRITE_TIME];
    }

    for (i = 0; i < total_io_ops_this_cycle; i++)
    {
        ind_coll_switch = (double)ind_io_ops_this_cycle / (total_io_ops_this_cycle - i);
        if (((double)rand() / (double)(RAND_MAX + 1.0)) < ind_coll_switch)
        {
            ind_coll = 0;
            tmp_rank = (next_ind_io_rank++) % nprocs;
            io_cnt = 1;
            if (!rw)
                file->counters[CP_INDEP_READS]--;
            else
                file->counters[CP_INDEP_WRITES]--;
            ind_io_ops_this_cycle--;
        }
        else
        {
            if (ind_ops_so_far)
            {
                cur_time = max_cur_time;
                cur_time += inter_io_delay;
                ind_ops_so_far = 0;
            }

            ind_coll = 1;
            cur_time = generate_barrier_event(file, 0, cur_time, io_context);
            tmp_rank = 0;
            if (!rw)
            {
                io_cnt = ceil((double)(file->counters[CP_POSIX_READS] -
                              file->counters[CP_INDEP_READS]) /
                              (file->counters[CP_COLL_READS] / nprocs));
                file->counters[CP_COLL_READS] -= nprocs;
            }
            else
            {
                io_cnt = ceil((double)(file->counters[CP_POSIX_WRITES] -
                              file->counters[CP_INDEP_WRITES]) /
                              (file->counters[CP_COLL_WRITES] / nprocs));
                file->counters[CP_COLL_WRITES] -= nprocs;
            }
            coll_io_ops_this_cycle--;
        }

        for (j = 0; j < io_cnt; j++)
        {
            determine_io_params(file, rw, ind_coll, (ind_coll) ? total_coll_io_ops :
                                file->counters[CP_POSIX_OPENS] / nprocs, &io_sz, &io_off);
            if (!rw)
            {
                /* generate a read event */
                next_io_op.codes_op.op_type = CODES_WK_READ;
                next_io_op.codes_op.u.read.file_id = file->hash;
                next_io_op.codes_op.u.read.size = io_sz;
                next_io_op.codes_op.u.read.offset = io_off;
                next_io_op.start_time = cur_time;

                /* set the end time based on observed bandwidth and io size */
                if (rd_bw == 0.0)
                    io_op_time = 0.0;
                else
                    io_op_time = (io_sz / rd_bw);
                
                next_io_op.end_time = cur_time + io_op_time + meta_op_time;
                file->counters[CP_POSIX_READS]--;
            }
            else
            {
                /* generate a write event */
                next_io_op.codes_op.op_type = CODES_WK_WRITE;
                next_io_op.codes_op.u.write.file_id = file->hash;
                next_io_op.codes_op.u.write.size = io_sz;
                next_io_op.codes_op.u.write.offset = io_off;
                next_io_op.start_time = cur_time;

                /* set the end time based on observed bandwidth and io size */
                if (wr_bw == 0.0)
                    io_op_time = 0.0;
                else
                    io_op_time = (io_sz / wr_bw);

                next_io_op.end_time = cur_time + io_op_time + meta_op_time;
                file->counters[CP_POSIX_WRITES]--;
            }
            psx_rw_ops_remaining--;
            assert(file->counters[CP_POSIX_READS] >= 0);
            assert(file->counters[CP_POSIX_WRITES] >= 0);

            /*  store the i/o event */
            if (tmp_rank == io_context->my_rank)
                darshan_insert_next_io_op(io_context->io_op_dat, &next_io_op);

            if (next_io_op.end_time > max_cur_time)
                max_cur_time = next_io_op.end_time;

            tmp_rank += ranks_per_aggregator;
            if (tmp_rank >= (ranks_per_aggregator * aggregator_cnt))
            {
                tmp_rank = 0;
                cur_time = max_cur_time;
                max_cur_time = 0.0;
            }
        }
        io_ops_this_rw--;

        /* determine whether to toggle between reads and writes */
        if (!io_ops_this_rw && psx_rw_ops_remaining)
        {
            /* toggle the read/write flag */
            rw ^= 1;
            file->counters[CP_RW_SWITCHES] -= aggregator_cnt;

            /* determine how many io ops to do before next rw switch */
            if (!rw)
            {
                if (file->counters[CP_COLL_OPENS])
                    io_ops_this_rw =
                        ((file->counters[CP_COLL_READS] / nprocs) +
                        file->counters[CP_INDEP_READS]) / ((file->counters[CP_RW_SWITCHES] /
                        (2 * aggregator_cnt)) + 1);
                else
                    io_ops_this_rw = file->counters[CP_POSIX_READS] /
                                     ((file->counters[CP_RW_SWITCHES] / (2 * aggregator_cnt)) + 1);
            }
            else
            {
                if (file->counters[CP_COLL_OPENS])
                    io_ops_this_rw =
                        ((file->counters[CP_COLL_WRITES] / nprocs) +
                        file->counters[CP_INDEP_WRITES]) / ((file->counters[CP_RW_SWITCHES] /
                        (2 * aggregator_cnt)) + 1);
                else
                    io_ops_this_rw = file->counters[CP_POSIX_WRITES] /
                                     ((file->counters[CP_RW_SWITCHES] / (2 * aggregator_cnt)) + 1);
            }
        }

        if (ind_coll)
        {
            total_coll_io_ops--;

            cur_time = max_cur_time;
            if (i != (total_io_ops_this_cycle - 1))
                cur_time += inter_io_delay;
        }
        else
        {
            if (ind_ops_so_far && ((ind_ops_so_far % nprocs) == nprocs - 1))
            {
                cur_time = max_cur_time;
                if (i != (total_io_ops_this_cycle - 1))
                    cur_time += inter_io_delay;
            }
            ind_ops_so_far++;
        }
    }

    /* reset the static rw flag if this is the last open-close cycle for this file */
    if (file->counters[CP_POSIX_OPENS] <= nprocs)
    {
        rw = -1;
    }

    if (max_cur_time > cur_time)
        return max_cur_time;
    else
        return cur_time;
}

1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
static void determine_io_params(
    struct darshan_file *file, int write_flag, int coll_flag, int64_t io_cycles, 
    size_t *io_sz, off_t *io_off)
{
    static int seq_rd_flag = -1;
    static int seq_wr_flag = -1;
    static uint64_t next_rd_off = 0;
    static uint64_t next_wr_off = 0;
    static int64_t rd_common_accesses[4] = { 0, 0, 0, 0 };
    static int64_t wr_common_accesses[4] = { 0, 0, 0, 0 };
    static int all_common_flag = -1;
    int64_t *size_bins; /* 10 size bins for io operations */
    int64_t *common_accesses = &(file->counters[CP_ACCESS1_ACCESS]); /* 4 common accesses */
    int64_t *common_access_counts = &(file->counters[CP_ACCESS1_COUNT]); /* common access counts */
    int64_t *total_io_size;
    int64_t last_io_byte;
    int64_t tmp_byte_counter = 0;
    int size_bin_ndx;
    int i, j = 0;
    int64_t bin_min_size[10] = { 0, 100, 1024, 10 * 1024, 100 * 1024, 1024 * 1024, 4 * 1024 * 1024,
                                 10 * 1024 * 1024, 100 * 1024 * 1024, 1024 * 1024 * 1024 };
    int64_t bin_def_size[10] = { 40, 512, 4 * 1024, 60 * 1024, 512 * 1024, 2 * 1024 * 1024,
                                 6 * 1024 * 1024, 40 * 1024 * 1024, 400 * 1024 * 1024,
                                 1 * 1024 * 1024 * 1024 };

    /* determine how to assign common access counters to reads and/or writes */
    if (all_common_flag == -1)
    {
        for (i = 0; i < 4; i++)
        {
            tmp_byte_counter += (common_accesses[i] * common_access_counts[i]);
        }

        if (tmp_byte_counter == (file->counters[CP_BYTES_WRITTEN] + file->counters[CP_BYTES_READ]))
        {
            all_common_flag = 1;
        }
        else
        {
            all_common_flag = 0;
        }
    }

    /* assign data values depending on whether the operation is a read or write */
    if (write_flag)
    {
        size_bins = &(file->counters[CP_SIZE_WRITE_0_100]);
        total_io_size = &(file->counters[CP_BYTES_WRITTEN]);
        last_io_byte = file->counters[CP_MAX_BYTE_WRITTEN];

        if (seq_wr_flag == -1)
        {
            if ((file->counters[CP_POSIX_WRITES] -
                ((*total_io_size - last_io_byte - 1) / (last_io_byte + 1)) - 1) ==
                file->counters[CP_SEQ_WRITES])
            {
                seq_wr_flag = 1;
            }
            else
            {
                seq_wr_flag = 0;
            }
        }
    }
    else
    {
        size_bins = &(file->counters[CP_SIZE_READ_0_100]);
        total_io_size = &(file->counters[CP_BYTES_READ]);
        last_io_byte = file->counters[CP_MAX_BYTE_READ];

        if (seq_rd_flag == -1)
        {
            if ((file->counters[CP_POSIX_READS] -
                ((*total_io_size - last_io_byte - 1) / (last_io_byte + 1)) - 1) ==
                file->counters[CP_SEQ_READS])
            {
                seq_rd_flag = 1;
            }
            else
            {
                seq_rd_flag = 0;
            }
        }
    }

    *io_sz = 0;
    if ((*total_io_size ==  0) || (write_flag && (file->counters[CP_POSIX_WRITES] == 1)) ||
        (!write_flag && (file->counters[CP_POSIX_READS] == 1)))
    {
        if (*total_io_size >= 0)
            *io_sz = *total_io_size;
    }
    else if (all_common_flag)
    {
        for (i = 0; i < 4; i++)
        {
            if (!write_flag && rd_common_accesses[i])
            {
                *io_sz = common_accesses[i];
                rd_common_accesses[i]--;
                common_access_counts[i]--;
                break;
            }
            else if (write_flag && wr_common_accesses[i])
            {
                *io_sz = common_accesses[i];
                wr_common_accesses[i]--;
                common_access_counts[i]--;
                break;
            }
        }

        if (*io_sz == 0)
        {
            for (i = 0; i < 4; i++)
            {
                if (write_flag)
                {
                    wr_common_accesses[i] = (common_access_counts[i] / io_cycles);
                    if ((*io_sz == 0) && wr_common_accesses[i])
                    {
                        *io_sz = common_accesses[i];
                        wr_common_accesses[i]--;
                        common_access_counts[i]--;
                    }
                }
                else
                {
                    rd_common_accesses[i] = (common_access_counts[i] / io_cycles);
                    if ((*io_sz == 0) && rd_common_accesses[i])
                    {
                        *io_sz = common_accesses[i];
                        rd_common_accesses[i]--;
                        common_access_counts[i]--;
                    }
                }
            }
        }
        assert(*io_sz);
    }
    else
    {
        /* try to assign a common access first */
        for (i = 0; i < 10; i++)
        {
            for (j = 0; j < 4; j++)
            {
                if (size_bins[i] && common_access_counts[j] &&
                    IO_IS_IN_SIZE_BIN_RANGE(common_accesses[j], i, bin_min_size))
                {
                    *io_sz = common_accesses[j];
                    common_access_counts[j]--;
                    break;
                }
            }
            if (*io_sz)
                break;
        }

        /* if no common accesses left, then assign a random io size */
        if (*io_sz == 0)
        {
            size_bin_ndx = rand() % 10;
            for (i = 0; i <  10; i++)
            {
                if (size_bins[size_bin_ndx])
                {
                    *io_sz = bin_def_size[size_bin_ndx];
                    break;
                }
                size_bin_ndx = (size_bin_ndx + 1) % 10;
            }
        }
        assert(*io_sz);
    }

    *total_io_size -= *io_sz;
    for (i = 0; i < 10; i++)
    {
        if (IO_IS_IN_SIZE_BIN_RANGE(*io_sz, i, bin_min_size))
            size_bins[i]--;
    }

    /* next, determine the offset to use */

    /*  for now we just assign a random offset that makes sure not to write past the recorded
     *  last byte written in the file.
     */
    if (*io_sz == 0)
    {
        *io_off = last_io_byte + 1;
    }
    else if (write_flag && seq_wr_flag)
    {
        if ((next_wr_off + *io_sz) > (last_io_byte + 1))
            next_wr_off = 0;

        *io_off = next_wr_off;
        next_wr_off += *io_sz;
    }
    else if (!write_flag && seq_rd_flag)
    {
        if ((next_rd_off + *io_sz) > (last_io_byte + 1))
            next_rd_off = 0;

        *io_off = next_rd_off;
        next_rd_off += *io_sz;
    }
    else if (*io_sz < last_io_byte)
    {
        *io_off = (off_t)rand() % (last_io_byte - *io_sz);
    }
    else
    {
        *io_off = 0;
    }

    /* reset static variable if this is the last i/o op for this file */
    if ((file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES]) == 1)
    {
        next_rd_off = next_wr_off = 0;
        seq_wr_flag = seq_rd_flag = -1;
        all_common_flag = -1;
        for (i = 0; i < 4; i++)
            rd_common_accesses[i] = wr_common_accesses[i] = 0;
    }

    return;
}

/* calculate the simulated "delay" between different i/o events using delay info
 * from the file counters */
static void calc_io_delays(
    struct darshan_file *file, int64_t num_opens, int64_t num_io_ops, double delay_per_cycle,
    double *first_io_delay, double *close_delay, double *inter_open_delay, double *inter_io_delay)
{
    double first_io_time, last_io_time;
    double first_io_pct, close_pct, inter_open_pct, inter_io_pct;
    double total_delay_pct;
    double tmp_inter_io_pct, tmp_inter_open_pct;

    if (delay_per_cycle > 0.0)
    {
        /* determine the time of the first io operation */
        if (!file->fcounters[CP_F_WRITE_START_TIMESTAMP])
            first_io_time = file->fcounters[CP_F_READ_START_TIMESTAMP];
        else if (!file->fcounters[CP_F_READ_START_TIMESTAMP])
            first_io_time = file->fcounters[CP_F_WRITE_START_TIMESTAMP];
        else if (file->fcounters[CP_F_READ_START_TIMESTAMP] <
                 file->fcounters[CP_F_WRITE_START_TIMESTAMP])
            first_io_time = file->fcounters[CP_F_READ_START_TIMESTAMP];
        else
            first_io_time = file->fcounters[CP_F_WRITE_START_TIMESTAMP];

        /* determine the time of the last io operation */
        if (file->fcounters[CP_F_READ_END_TIMESTAMP] > file->fcounters[CP_F_WRITE_END_TIMESTAMP])
            last_io_time = file->fcounters[CP_F_READ_END_TIMESTAMP];
        else
            last_io_time = file->fcounters[CP_F_WRITE_END_TIMESTAMP];

        /* no delay contribution for inter-open delay if there is only a single open */
        if (num_opens > 1)
            inter_open_pct = DEF_INTER_CYC_DELAY_PCT;

        /* no delay contribution for inter-io delay if there is one or less io op */
1458
        if ((num_io_ops - num_opens) > 0)
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
            inter_io_pct = DEF_INTER_IO_DELAY_PCT;

        /* determine delay contribution for first io and close delays */
        if (first_io_time != 0.0)
        {
            first_io_pct = (first_io_time - file->fcounters[CP_F_OPEN_TIMESTAMP]) / delay_per_cycle;
            close_pct = (file->fcounters[CP_F_CLOSE_TIMESTAMP] - last_io_time) / delay_per_cycle;
        }
        else
        {
            first_io_pct = 0.0;
            close_pct = 1 - inter_open_pct;
        }

        /* adjust per open delay percentages using a simple heuristic */
        total_delay_pct = inter_open_pct + inter_io_pct + first_io_pct + close_pct;
        if ((total_delay_pct < 1) && (inter_open_pct || inter_io_pct))
        {
            /* only adjust inter-open and inter-io delays if we underestimate */
            tmp_inter_open_pct = (inter_open_pct / (inter_open_pct + inter_io_pct)) *
                                 (1 - first_io_pct - close_pct);
            tmp_inter_io_pct = (inter_io_pct / (inter_open_pct + inter_io_pct)) *
                               (1 - first_io_pct - close_pct);
            inter_open_pct = tmp_inter_open_pct;
            inter_io_pct = tmp_inter_io_pct;
        }
        else
        {
            inter_open_pct += (inter_open_pct / total_delay_pct) * (1 - total_delay_pct);
            inter_io_pct += (inter_io_pct / total_delay_pct) * (1 - total_delay_pct);
            first_io_pct += (first_io_pct / total_delay_pct) * (1 - total_delay_pct);
            close_pct += (close_pct / total_delay_pct) * (1 - total_delay_pct);
        }

        *first_io_delay = (first_io_pct * delay_per_cycle);
        *close_delay = (close_pct * delay_per_cycle);
1495
1496
1497
1498
1499
1500
1501

        if (num_opens > 1)
            *inter_open_delay = (inter_open_pct * delay_per_cycle) *
                                ((double)num_opens / (num_opens - 1));
        if ((num_io_ops - num_opens) > 0)
            *inter_io_delay = (inter_io_pct * delay_per_cycle) *
                              ((double)num_io_ops / (num_io_ops - num_opens));
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
    }

    return;
}

/* check to make sure file stats are valid and properly formatted */
static void file_sanity_check(
    struct darshan_file *file, struct darshan_job *job)
{
    assert(file->counters[CP_POSIX_OPENS] != -1);
    assert(file->fcounters[CP_F_OPEN_TIMESTAMP] != -1);
    assert(file->counters[CP_COLL_OPENS] != -1);
    assert(file->fcounters[CP_F_CLOSE_TIMESTAMP] != -1);
    assert(file->counters[CP_POSIX_READS] != -1);
    assert(file->counters[CP_POSIX_WRITES] != -1);
    assert(file->fcounters[CP_F_POSIX_READ_TIME] != -1);
    assert(file->fcounters[CP_F_POSIX_WRITE_TIME] != -1);
    assert(file->fcounters[CP_F_POSIX_META_TIME] != -1);
    assert(file->fcounters[CP_F_READ_START_TIMESTAMP] != -1);
    assert(file->fcounters[CP_F_WRITE_START_TIMESTAMP] != -1);
    assert(file->fcounters[CP_F_READ_END_TIMESTAMP] != -1);
    assert(file->fcounters[CP_F_WRITE_END_TIMESTAMP] != -1);
    assert(file->counters[CP_BYTES_READ] != -1);
    assert(file->counters[CP_BYTES_WRITTEN] != -1);
    assert(file->counters[CP_RW_SWITCHES] != -1);

    /* adjust timestamps if they are given in absolute unix time */
    if (file->fcounters[CP_F_OPEN_TIMESTAMP] > job->start_time)
    {
        file->fcounters[CP_F_OPEN_TIMESTAMP] -= job->start_time;
        if (file->fcounters[CP_F_OPEN_TIMESTAMP] < 0.0)
            file->fcounters[CP_F_OPEN_TIMESTAMP] = 0.0;

        file->fcounters[CP_F_READ_START_TIMESTAMP] -= job->start_time;
        if (file->fcounters[CP_F_READ_START_TIMESTAMP] < 0.0)
            file->fcounters[CP_F_READ_START_TIMESTAMP] = 0.0;

        file->fcounters[CP_F_WRITE_START_TIMESTAMP] -= job->start_time;
        if (file->fcounters[CP_F_WRITE_START_TIMESTAMP] < 0.0)
            file->fcounters[CP_F_WRITE_START_TIMESTAMP] = 0.0;

        file->fcounters[CP_F_CLOSE_TIMESTAMP] -= job->start_time;
        if (file->fcounters[CP_F_CLOSE_TIMESTAMP] < 0.0)
            file->fcounters[CP_F_CLOSE_TIMESTAMP] = 0.0;

        file->fcounters[CP_F_READ_END_TIMESTAMP] -= job->start_time;
        if (file->fcounters[CP_F_READ_END_TIMESTAMP] < 0.0)
            file->fcounters[CP_F_READ_END_TIMESTAMP] = 0.0;

        file->fcounters[CP_F_WRITE_END_TIMESTAMP] -= job->start_time;
        if (file->fcounters[CP_F_WRITE_END_TIMESTAMP] < 0.0)
            file->fcounters[CP_F_WRITE_END_TIMESTAMP] = 0.0;
    }

    /* set file close time to the end of execution if it is not given */
    if (file->fcounters[CP_F_CLOSE_TIMESTAMP] == 0.0)
        file->fcounters[CP_F_CLOSE_TIMESTAMP] = job->end_time - job->start_time + 1;

    /* collapse fopen/fread/etc. calls into the corresponding open/read/etc. counters */
    file->counters[CP_POSIX_OPENS] += file->counters[CP_POSIX_FOPENS];
    file->counters[CP_POSIX_READS] += file->counters[CP_POSIX_FREADS];
    file->counters[CP_POSIX_WRITES] += file->counters[CP_POSIX_FWRITES];

    return;
}
1567
1568
1569
1570
1571
1572
1573
1574
1575

/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ft=c ts=8 sts=4 sw=4 expandtab
 */