14 void update_rusage_stat(struct thread_data *td)
16 struct thread_stat *ts = &td->ts;
18 getrusage(RUSAGE_SELF, &td->ru_end);
20 ts->usr_time += mtime_since(&td->ru_start.ru_utime,
21 &td->ru_end.ru_utime);
22 ts->sys_time += mtime_since(&td->ru_start.ru_stime,
23 &td->ru_end.ru_stime);
24 ts->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw
25 - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw);
26 ts->minf += td->ru_end.ru_minflt - td->ru_start.ru_minflt;
27 ts->majf += td->ru_end.ru_majflt - td->ru_start.ru_majflt;
29 memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end));
33 * Given a latency, return the index of the corresponding bucket in
34 * the structure tracking percentiles.
36 * (1) find the group (and error bits) that the value (latency)
37 * belongs to by looking at its MSB. (2) find the bucket number in the
38 * group by looking at the index bits.
41 static unsigned int plat_val_to_idx(unsigned int val)
43 unsigned int msb, error_bits, base, offset, idx;
45 /* Find MSB starting from bit 0 */
49 msb = (sizeof(val)*8) - __builtin_clz(val) - 1;
52 * MSB <= (FIO_IO_U_PLAT_BITS-1), cannot be rounded off. Use
53 * all bits of the sample as index
55 if (msb <= FIO_IO_U_PLAT_BITS)
58 /* Compute the number of error bits to discard*/
59 error_bits = msb - FIO_IO_U_PLAT_BITS;
61 /* Compute the number of buckets before the group */
62 base = (error_bits + 1) << FIO_IO_U_PLAT_BITS;
65 * Discard the error bits and apply the mask to find the
66 * index for the buckets in the group
68 offset = (FIO_IO_U_PLAT_VAL - 1) & (val >> error_bits);
70 /* Make sure the index does not exceed (array size - 1) */
71 idx = (base + offset) < (FIO_IO_U_PLAT_NR - 1)?
72 (base + offset) : (FIO_IO_U_PLAT_NR - 1);
78 * Convert the given index of the bucket array to the value
79 * represented by the bucket
81 static unsigned int plat_idx_to_val(unsigned int idx)
83 unsigned int error_bits, k, base;
85 assert(idx < FIO_IO_U_PLAT_NR);
87 /* MSB <= (FIO_IO_U_PLAT_BITS-1), cannot be rounded off. Use
88 * all bits of the sample as index */
89 if (idx < (FIO_IO_U_PLAT_VAL << 1) )
92 /* Find the group and compute the minimum value of that group */
93 error_bits = (idx >> FIO_IO_U_PLAT_BITS) -1;
94 base = 1 << (error_bits + FIO_IO_U_PLAT_BITS);
96 /* Find its bucket number of the group */
97 k = idx % FIO_IO_U_PLAT_VAL;
99 /* Return the mean of the range of the bucket */
100 return base + ((k + 0.5) * (1 << error_bits));
103 static int double_cmp(const void *a, const void *b)
105 const fio_fp64_t fa = *(const fio_fp64_t *) a;
106 const fio_fp64_t fb = *(const fio_fp64_t *) b;
111 else if (fa.u.f < fb.u.f)
118 * Find and display the p-th percentile of clat
120 static void show_clat_percentiles(unsigned int *io_u_plat, unsigned long nr,
123 unsigned long sum = 0;
124 unsigned int len, i, j = 0;
128 while (len < FIO_IO_U_LIST_MAX_LEN && plist[len].u.f != 0.0)
132 * Sort the percentile list. Note that it may already be sorted if
133 * we are using the default values, but since it's a short list this
134 * isn't a worry. Also note that this does not work for NaN values.
137 qsort((void*)plist, len, sizeof(plist[0]), double_cmp);
139 log_info(" clat percentiles (usec) :");
141 for (i = 0; i < FIO_IO_U_PLAT_NR && !is_last; i++) {
143 while (sum >= (plist[j].u.f / 100.0 * nr)) {
144 assert(plist[j].u.f <= 100.0);
147 if (j != 0 && (j % 4) == 0)
150 /* end of the list */
151 is_last = (j == len - 1);
153 log_info(" %2.2fth=%u%c", plist[j], plat_idx_to_val(i),
154 (is_last? '\n' : ','));
159 if (j % 4 == 3) /* for formatting */
166 static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
167 double *mean, double *dev)
169 double n = is->samples;
171 if (is->samples == 0)
177 n = (double) is->samples;
178 *mean = is->mean.u.f;
181 *dev = sqrt(is->S.u.f / (n - 1.0));
188 void show_group_stats(struct group_run_stats *rs)
190 char *p1, *p2, *p3, *p4;
191 const char *ddir_str[] = { " READ", " WRITE" };
194 log_info("\nRun status group %d (all jobs):\n", rs->groupid);
196 for (i = 0; i <= DDIR_WRITE; i++) {
197 const int i2p = is_power_of_2(rs->kb_base);
202 p1 = num2str(rs->io_kb[i], 6, rs->kb_base, i2p);
203 p2 = num2str(rs->agg[i], 6, rs->kb_base, i2p);
204 p3 = num2str(rs->min_bw[i], 6, rs->kb_base, i2p);
205 p4 = num2str(rs->max_bw[i], 6, rs->kb_base, i2p);
207 log_info("%s: io=%sB, aggrb=%sB/s, minb=%sB/s, maxb=%sB/s,"
208 " mint=%llumsec, maxt=%llumsec\n", ddir_str[i], p1, p2,
209 p3, p4, rs->min_run[i],
219 #define ts_total_io_u(ts) \
220 ((ts)->total_io_u[0] + (ts)->total_io_u[1])
222 static void stat_calc_dist(unsigned int *map, unsigned long total,
228 * Do depth distribution calculations
230 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
232 io_u_dist[i] = (double) map[i] / (double) total;
233 io_u_dist[i] *= 100.0;
234 if (io_u_dist[i] < 0.1 && map[i])
241 static void stat_calc_lat(struct thread_stat *ts, double *dst,
242 unsigned int *src, int nr)
244 unsigned long total = ts_total_io_u(ts);
248 * Do latency distribution calculations
250 for (i = 0; i < nr; i++) {
252 dst[i] = (double) src[i] / (double) total;
254 if (dst[i] < 0.01 && src[i])
261 static void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat)
263 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_u, FIO_IO_U_LAT_U_NR);
266 static void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat)
268 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_m, FIO_IO_U_LAT_M_NR);
271 static int usec_to_msec(unsigned long *min, unsigned long *max, double *mean,
274 if (*min > 1000 && *max > 1000 && *mean > 1000.0 && *dev > 1000.0) {
285 static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
288 const char *ddir_str[] = { "read ", "write" };
289 unsigned long min, max, runt;
290 unsigned long long bw, iops;
292 char *io_p, *bw_p, *iops_p;
295 assert(ddir_rw(ddir));
297 if (!ts->runtime[ddir])
300 i2p = is_power_of_2(rs->kb_base);
301 runt = ts->runtime[ddir];
303 bw = (1000 * ts->io_bytes[ddir]) / runt;
304 io_p = num2str(ts->io_bytes[ddir], 6, 1, i2p);
305 bw_p = num2str(bw, 6, 1, i2p);
307 iops = (1000 * (uint64_t)ts->total_io_u[ddir]) / runt;
308 iops_p = num2str(iops, 6, 1, 0);
310 log_info(" %s: io=%sB, bw=%sB/s, iops=%s, runt=%6llumsec\n",
311 ddir_str[ddir], io_p, bw_p, iops_p,
318 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) {
319 const char *base = "(usec)";
322 if (!usec_to_msec(&min, &max, &mean, &dev))
325 minp = num2str(min, 6, 1, 0);
326 maxp = num2str(max, 6, 1, 0);
328 log_info(" slat %s: min=%s, max=%s, avg=%5.02f,"
329 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
334 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) {
335 const char *base = "(usec)";
338 if (!usec_to_msec(&min, &max, &mean, &dev))
341 minp = num2str(min, 6, 1, 0);
342 maxp = num2str(max, 6, 1, 0);
344 log_info(" clat %s: min=%s, max=%s, avg=%5.02f,"
345 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
350 if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev)) {
351 const char *base = "(usec)";
354 if (!usec_to_msec(&min, &max, &mean, &dev))
357 minp = num2str(min, 6, 1, 0);
358 maxp = num2str(max, 6, 1, 0);
360 log_info(" lat %s: min=%s, max=%s, avg=%5.02f,"
361 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
366 if (ts->clat_percentiles) {
367 show_clat_percentiles(ts->io_u_plat[ddir],
368 ts->clat_stat[ddir].samples,
369 ts->percentile_list);
371 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
374 p_of_agg = mean * 100 / (double) rs->agg[ddir];
375 log_info(" bw (KB/s) : min=%5lu, max=%5lu, per=%3.2f%%,"
376 " avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg,
381 static void show_lat(double *io_u_lat, int nr, const char **ranges,
384 int new_line = 1, i, line = 0;
386 for (i = 0; i < nr; i++) {
387 if (io_u_lat[i] <= 0.0)
392 log_info(" lat (%s): ", msg);
398 log_info("%s%3.2f%%", ranges[i], io_u_lat[i]);
405 static void show_lat_u(double *io_u_lat_u)
407 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
408 "250=", "500=", "750=", "1000=", };
410 show_lat(io_u_lat_u, FIO_IO_U_LAT_U_NR, ranges, "usec");
413 static void show_lat_m(double *io_u_lat_m)
415 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
416 "250=", "500=", "750=", "1000=", "2000=",
419 show_lat(io_u_lat_m, FIO_IO_U_LAT_M_NR, ranges, "msec");
422 static void show_latencies(double *io_u_lat_u, double *io_u_lat_m)
424 show_lat_u(io_u_lat_u);
426 show_lat_m(io_u_lat_m);
430 void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs)
432 double usr_cpu, sys_cpu;
433 unsigned long runtime;
434 double io_u_dist[FIO_IO_U_MAP_NR];
435 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
436 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
438 if (!(ts->io_bytes[0] + ts->io_bytes[1]) &&
439 !(ts->total_io_u[0] + ts->total_io_u[1]))
443 log_info("%s: (groupid=%d, jobs=%d): err=%2d: pid=%d\n",
444 ts->name, ts->groupid, ts->members,
445 ts->error, (int) ts->pid);
447 log_info("%s: (groupid=%d, jobs=%d): err=%2d (%s): pid=%d\n",
448 ts->name, ts->groupid, ts->members,
449 ts->error, ts->verror, (int) ts->pid);
453 log_info(" Description : [%s]\n", ts->description);
455 if (ts->io_bytes[DDIR_READ])
456 show_ddir_status(rs, ts, DDIR_READ);
457 if (ts->io_bytes[DDIR_WRITE])
458 show_ddir_status(rs, ts, DDIR_WRITE);
460 runtime = ts->total_run_time;
462 double runt = (double) runtime;
464 usr_cpu = (double) ts->usr_time * 100 / runt;
465 sys_cpu = (double) ts->sys_time * 100 / runt;
471 log_info(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu, majf=%lu,"
472 " minf=%lu\n", usr_cpu, sys_cpu, ts->ctx, ts->majf, ts->minf);
474 stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
475 log_info(" IO depths : 1=%3.1f%%, 2=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%,"
476 " 16=%3.1f%%, 32=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
477 io_u_dist[1], io_u_dist[2],
478 io_u_dist[3], io_u_dist[4],
479 io_u_dist[5], io_u_dist[6]);
481 stat_calc_dist(ts->io_u_submit, ts->total_submit, io_u_dist);
482 log_info(" submit : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
483 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
484 io_u_dist[1], io_u_dist[2],
485 io_u_dist[3], io_u_dist[4],
486 io_u_dist[5], io_u_dist[6]);
487 stat_calc_dist(ts->io_u_complete, ts->total_complete, io_u_dist);
488 log_info(" complete : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
489 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
490 io_u_dist[1], io_u_dist[2],
491 io_u_dist[3], io_u_dist[4],
492 io_u_dist[5], io_u_dist[6]);
493 log_info(" issued r/w/d: total=%lu/%lu/%lu, short=%lu/%lu/%lu\n",
494 ts->total_io_u[0], ts->total_io_u[1],
496 ts->short_io_u[0], ts->short_io_u[1],
498 stat_calc_lat_u(ts, io_u_lat_u);
499 stat_calc_lat_m(ts, io_u_lat_m);
500 show_latencies(io_u_lat_u, io_u_lat_m);
501 if (ts->continue_on_error) {
502 log_info(" errors : total=%lu, first_error=%d/<%s>\n",
505 strerror(ts->first_error));
509 static void show_ddir_status_terse(struct thread_stat *ts,
510 struct group_run_stats *rs, int ddir)
512 unsigned long min, max;
513 unsigned long long bw;
516 assert(ddir_rw(ddir));
519 if (ts->runtime[ddir])
520 bw = ts->io_bytes[ddir] / ts->runtime[ddir];
522 log_info(";%llu;%llu;%llu", ts->io_bytes[ddir] >> 10, bw,
525 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
526 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
528 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
530 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
531 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
533 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
535 if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev))
536 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
538 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
540 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
543 p_of_agg = mean * 100 / (double) rs->agg[ddir];
544 log_info(";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
546 log_info(";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0);
549 #define FIO_TERSE_VERSION "2"
551 static void show_thread_status_terse(struct thread_stat *ts,
552 struct group_run_stats *rs)
554 double io_u_dist[FIO_IO_U_MAP_NR];
555 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
556 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
557 double usr_cpu, sys_cpu;
561 log_info("%s;%s;%d;%d", FIO_TERSE_VERSION, ts->name, ts->groupid,
563 /* Log Read Status */
564 show_ddir_status_terse(ts, rs, 0);
565 /* Log Write Status */
566 show_ddir_status_terse(ts, rs, 1);
569 if (ts->total_run_time) {
570 double runt = (double) ts->total_run_time;
572 usr_cpu = (double) ts->usr_time * 100 / runt;
573 sys_cpu = (double) ts->sys_time * 100 / runt;
579 log_info(";%f%%;%f%%;%lu;%lu;%lu", usr_cpu, sys_cpu, ts->ctx, ts->majf,
582 /* Calc % distribution of IO depths, usecond, msecond latency */
583 stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
584 stat_calc_lat_u(ts, io_u_lat_u);
585 stat_calc_lat_m(ts, io_u_lat_m);
587 /* Only show fixed 7 I/O depth levels*/
588 log_info(";%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%",
589 io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3],
590 io_u_dist[4], io_u_dist[5], io_u_dist[6]);
592 /* Microsecond latency */
593 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
594 log_info(";%3.2f%%", io_u_lat_u[i]);
595 /* Millisecond latency */
596 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
597 log_info(";%3.2f%%", io_u_lat_m[i]);
598 /* Additional output if continue_on_error set - default off*/
599 if (ts->continue_on_error)
600 log_info(";%lu;%d", ts->total_err_count, ts->first_error);
603 /* Additional output if description is set */
605 log_info(";%s", ts->description);
610 static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr)
614 if (src->samples == 0)
617 dst->min_val = min(dst->min_val, src->min_val);
618 dst->max_val = max(dst->max_val, src->max_val);
621 * Compute new mean and S after the merge
622 * <http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
623 * #Parallel_algorithm>
626 mean = src->mean.u.f;
629 double delta = src->mean.u.f - dst->mean.u.f;
631 mean = ((src->mean.u.f * src->samples) +
632 (dst->mean.u.f * dst->samples)) /
633 (dst->samples + src->samples);
635 S = src->S.u.f + dst->S.u.f + pow(delta, 2.0) *
636 (dst->samples * src->samples) /
637 (dst->samples + src->samples);
640 dst->samples += src->samples;
641 dst->mean.u.f = mean;
645 void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src)
649 for (i = 0; i < 2; i++) {
650 if (dst->max_run[i] < src->max_run[i])
651 dst->max_run[i] = src->max_run[i];
652 if (dst->min_run[i] && dst->min_run[i] > src->min_run[i])
653 dst->min_run[i] = src->min_run[i];
654 if (dst->max_bw[i] < src->max_bw[i])
655 dst->max_bw[i] = src->max_bw[i];
656 if (dst->min_bw[i] && dst->min_bw[i] > src->min_bw[i])
657 dst->min_bw[i] = src->min_bw[i];
659 dst->io_kb[i] += src->io_kb[i];
660 dst->agg[i] += src->agg[i];
665 void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, int nr)
669 for (l = 0; l <= DDIR_WRITE; l++) {
670 sum_stat(&dst->clat_stat[l], &src->clat_stat[l], nr);
671 sum_stat(&dst->slat_stat[l], &src->slat_stat[l], nr);
672 sum_stat(&dst->lat_stat[l], &src->lat_stat[l], nr);
673 sum_stat(&dst->bw_stat[l], &src->bw_stat[l], nr);
675 dst->io_bytes[l] += src->io_bytes[l];
677 if (dst->runtime[l] < src->runtime[l])
678 dst->runtime[l] = src->runtime[l];
681 dst->usr_time += src->usr_time;
682 dst->sys_time += src->sys_time;
683 dst->ctx += src->ctx;
684 dst->majf += src->majf;
685 dst->minf += src->minf;
687 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
688 dst->io_u_map[k] += src->io_u_map[k];
689 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
690 dst->io_u_submit[k] += src->io_u_submit[k];
691 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
692 dst->io_u_complete[k] += src->io_u_complete[k];
693 for (k = 0; k < FIO_IO_U_LAT_U_NR; k++)
694 dst->io_u_lat_u[k] += src->io_u_lat_u[k];
695 for (k = 0; k < FIO_IO_U_LAT_M_NR; k++)
696 dst->io_u_lat_m[k] += src->io_u_lat_m[k];
698 for (k = 0; k <= 2; k++) {
699 dst->total_io_u[k] += src->total_io_u[k];
700 dst->short_io_u[k] += src->short_io_u[k];
703 for (k = 0; k <= DDIR_WRITE; k++) {
705 for (m = 0; m < FIO_IO_U_PLAT_NR; m++)
706 dst->io_u_plat[k][m] += src->io_u_plat[k][m];
709 dst->total_run_time += src->total_run_time;
710 dst->total_submit += src->total_submit;
711 dst->total_complete += src->total_complete;
714 void init_group_run_stat(struct group_run_stats *gs)
716 memset(gs, 0, sizeof(*gs));
717 gs->min_bw[0] = gs->min_run[0] = ~0UL;
718 gs->min_bw[1] = gs->min_run[1] = ~0UL;
721 void init_thread_stat(struct thread_stat *ts)
725 memset(ts, 0, sizeof(*ts));
727 for (j = 0; j <= DDIR_WRITE; j++) {
728 ts->lat_stat[j].min_val = -1UL;
729 ts->clat_stat[j].min_val = -1UL;
730 ts->slat_stat[j].min_val = -1UL;
731 ts->bw_stat[j].min_val = -1UL;
736 void show_run_stats(void)
738 struct group_run_stats *runstats, *rs;
739 struct thread_data *td;
740 struct thread_stat *threadstats, *ts;
741 int i, j, nr_ts, last_ts, idx;
742 int kb_base_warned = 0;
744 runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
746 for (i = 0; i < groupid + 1; i++)
747 init_group_run_stat(&runstats[i]);
750 * find out how many threads stats we need. if group reporting isn't
751 * enabled, it's one-per-td.
756 if (!td->o.group_reporting) {
760 if (last_ts == td->groupid)
763 last_ts = td->groupid;
767 threadstats = malloc(nr_ts * sizeof(struct thread_stat));
769 for (i = 0; i < nr_ts; i++)
770 init_thread_stat(&threadstats[i]);
776 if (idx && (!td->o.group_reporting ||
777 (td->o.group_reporting && last_ts != td->groupid))) {
782 last_ts = td->groupid;
784 ts = &threadstats[j];
786 ts->clat_percentiles = td->o.clat_percentiles;
787 if (td->o.overwrite_plist)
788 memcpy(ts->percentile_list, td->o.percentile_list, sizeof(td->o.percentile_list));
790 memcpy(ts->percentile_list, def_percentile_list, sizeof(def_percentile_list));
795 if (ts->groupid == -1) {
797 * These are per-group shared already
799 strncpy(ts->name, td->o.name, FIO_JOBNAME_SIZE);
800 if (td->o.description)
801 strncpy(ts->description, td->o.description,
804 memset(ts->description, 0, FIO_JOBNAME_SIZE);
806 ts->groupid = td->groupid;
809 * first pid in group, not very useful...
813 ts->kb_base = td->o.kb_base;
814 } else if (ts->kb_base != td->o.kb_base && !kb_base_warned) {
815 log_info("fio: kb_base differs for jobs in group, using"
816 " %u as the base\n", ts->kb_base);
820 ts->continue_on_error = td->o.continue_on_error;
821 ts->total_err_count += td->total_err_count;
822 ts->first_error = td->first_error;
824 if (!td->error && td->o.continue_on_error &&
826 ts->error = td->first_error;
827 strcpy(ts->verror, td->verror);
828 } else if (td->error) {
829 ts->error = td->error;
830 strcpy(ts->verror, td->verror);
834 sum_thread_stats(ts, &td->ts, idx);
837 for (i = 0; i < nr_ts; i++) {
838 unsigned long long bw;
840 ts = &threadstats[i];
841 rs = &runstats[ts->groupid];
842 rs->kb_base = ts->kb_base;
844 for (j = 0; j <= DDIR_WRITE; j++) {
847 if (ts->runtime[j] < rs->min_run[j] || !rs->min_run[j])
848 rs->min_run[j] = ts->runtime[j];
849 if (ts->runtime[j] > rs->max_run[j])
850 rs->max_run[j] = ts->runtime[j];
853 if (ts->runtime[j]) {
856 runt = ts->runtime[j];
857 bw = ts->io_bytes[j] / runt;
859 if (bw < rs->min_bw[j])
861 if (bw > rs->max_bw[j])
864 rs->io_kb[j] += ts->io_bytes[j] / rs->kb_base;
868 for (i = 0; i < groupid + 1; i++) {
869 unsigned long max_run[2];
872 max_run[0] = rs->max_run[0];
873 max_run[1] = rs->max_run[1];
876 rs->agg[0] = (rs->io_kb[0] * 1000) / max_run[0];
878 rs->agg[1] = (rs->io_kb[1] * 1000) / max_run[1];
882 * don't overwrite last signal output
887 for (i = 0; i < nr_ts; i++) {
888 ts = &threadstats[i];
889 rs = &runstats[ts->groupid];
892 fio_server_send_ts(ts, rs);
893 else if (terse_output)
894 show_thread_status_terse(ts, rs);
896 show_thread_status(ts, rs);
900 for (i = 0; i < groupid + 1; i++) {
905 fio_server_send_gs(rs);
907 show_group_stats(rs);
917 static inline void add_stat_sample(struct io_stat *is, unsigned long data)
922 if (data > is->max_val)
924 if (data < is->min_val)
927 delta = val - is->mean.u.f;
929 is->mean.u.f += delta / (is->samples + 1.0);
930 is->S.u.f += delta * (val - is->mean.u.f);
936 static void __add_log_sample(struct io_log *iolog, unsigned long val,
937 enum fio_ddir ddir, unsigned int bs,
940 const int nr_samples = iolog->nr_samples;
942 if (iolog->nr_samples == iolog->max_samples) {
943 int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
945 iolog->log = realloc(iolog->log, new_size);
946 iolog->max_samples <<= 1;
949 iolog->log[nr_samples].val = val;
950 iolog->log[nr_samples].time = t;
951 iolog->log[nr_samples].ddir = ddir;
952 iolog->log[nr_samples].bs = bs;
956 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
957 unsigned long val, enum fio_ddir ddir,
963 __add_log_sample(iolog, val, ddir, bs, mtime_since_now(&td->epoch));
966 void add_agg_sample(unsigned long val, enum fio_ddir ddir, unsigned int bs)
968 struct io_log *iolog;
973 iolog = agg_io_log[ddir];
974 __add_log_sample(iolog, val, ddir, bs, mtime_since_genesis());
977 static void add_clat_percentile_sample(struct thread_stat *ts,
978 unsigned long usec, enum fio_ddir ddir)
980 unsigned int idx = plat_val_to_idx(usec);
981 assert(idx < FIO_IO_U_PLAT_NR);
983 ts->io_u_plat[ddir][idx]++;
986 void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
987 unsigned long usec, unsigned int bs)
989 struct thread_stat *ts = &td->ts;
994 add_stat_sample(&ts->clat_stat[ddir], usec);
997 add_log_sample(td, td->clat_log, usec, ddir, bs);
999 if (ts->clat_percentiles)
1000 add_clat_percentile_sample(ts, usec, ddir);
1003 void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
1004 unsigned long usec, unsigned int bs)
1006 struct thread_stat *ts = &td->ts;
1011 add_stat_sample(&ts->slat_stat[ddir], usec);
1014 add_log_sample(td, td->slat_log, usec, ddir, bs);
1017 void add_lat_sample(struct thread_data *td, enum fio_ddir ddir,
1018 unsigned long usec, unsigned int bs)
1020 struct thread_stat *ts = &td->ts;
1025 add_stat_sample(&ts->lat_stat[ddir], usec);
1028 add_log_sample(td, td->lat_log, usec, ddir, bs);
1031 void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs,
1034 struct thread_stat *ts = &td->ts;
1035 unsigned long spent, rate;
1040 spent = mtime_since(&td->bw_sample_time, t);
1041 if (spent < td->o.bw_avg_time)
1044 rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) *
1045 1000 / spent / 1024;
1046 add_stat_sample(&ts->bw_stat[ddir], rate);
1049 add_log_sample(td, td->bw_log, rate, ddir, bs);
1051 fio_gettime(&td->bw_sample_time, NULL);
1052 td->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
1055 void add_iops_sample(struct thread_data *td, enum fio_ddir ddir,
1058 struct thread_stat *ts = &td->ts;
1059 unsigned long spent, iops;
1064 spent = mtime_since(&td->iops_sample_time, t);
1065 if (spent < td->o.iops_avg_time)
1068 iops = ((td->this_io_blocks[ddir] - td->stat_io_blocks[ddir]) * 1000) / spent;
1070 add_stat_sample(&ts->iops_stat[ddir], iops);
1074 add_log_sample(td, td->iops_log, iops, ddir, 0);
1077 fio_gettime(&td->iops_sample_time, NULL);
1078 td->stat_io_blocks[ddir] = td->this_io_blocks[ddir];