13 void update_rusage_stat(struct thread_data *td)
15 struct thread_stat *ts = &td->ts;
17 getrusage(RUSAGE_SELF, &ts->ru_end);
19 ts->usr_time += mtime_since(&ts->ru_start.ru_utime,
20 &ts->ru_end.ru_utime);
21 ts->sys_time += mtime_since(&ts->ru_start.ru_stime,
22 &ts->ru_end.ru_stime);
23 ts->ctx += ts->ru_end.ru_nvcsw + ts->ru_end.ru_nivcsw
24 - (ts->ru_start.ru_nvcsw + ts->ru_start.ru_nivcsw);
25 ts->minf += ts->ru_end.ru_minflt - ts->ru_start.ru_minflt;
26 ts->majf += ts->ru_end.ru_majflt - ts->ru_start.ru_majflt;
28 memcpy(&ts->ru_start, &ts->ru_end, sizeof(ts->ru_end));
31 static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
32 double *mean, double *dev)
34 double n = is->samples;
42 n = (double) is->samples;
46 *dev = sqrt(is->S / (n - 1.0));
53 static void show_group_stats(struct group_run_stats *rs, int id)
55 char *p1, *p2, *p3, *p4;
56 const char *ddir_str[] = { " READ", " WRITE" };
59 log_info("\nRun status group %d (all jobs):\n", id);
61 for (i = 0; i <= DDIR_WRITE; i++) {
62 const int i2p = is_power_of_2(rs->kb_base);
67 p1 = num2str(rs->io_kb[i], 6, rs->kb_base, i2p);
68 p2 = num2str(rs->agg[i], 6, rs->kb_base, i2p);
69 p3 = num2str(rs->min_bw[i], 6, rs->kb_base, i2p);
70 p4 = num2str(rs->max_bw[i], 6, rs->kb_base, i2p);
72 log_info("%s: io=%sB, aggrb=%sB/s, minb=%sB/s, maxb=%sB/s,"
73 " mint=%llumsec, maxt=%llumsec\n", ddir_str[i], p1, p2,
74 p3, p4, rs->min_run[i],
84 #define ts_total_io_u(ts) \
85 ((ts)->total_io_u[0] + (ts)->total_io_u[1])
87 static void stat_calc_dist(unsigned int *map, unsigned long total,
93 * Do depth distribution calculations
95 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
97 io_u_dist[i] = (double) map[i] / (double) total;
98 io_u_dist[i] *= 100.0;
99 if (io_u_dist[i] < 0.1 && map[i])
106 static void stat_calc_lat(struct thread_stat *ts, double *dst,
107 unsigned int *src, int nr)
109 unsigned long total = ts_total_io_u(ts);
113 * Do latency distribution calculations
115 for (i = 0; i < nr; i++) {
117 dst[i] = (double) src[i] / (double) total;
119 if (dst[i] < 0.01 && src[i])
126 static void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat)
128 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_u, FIO_IO_U_LAT_U_NR);
131 static void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat)
133 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_m, FIO_IO_U_LAT_M_NR);
136 static int usec_to_msec(unsigned long *min, unsigned long *max, double *mean,
139 if (*min > 1000 && *max > 1000 && *mean > 1000.0 && *dev > 1000.0) {
150 static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
153 const char *ddir_str[] = { "read ", "write" };
154 unsigned long min, max, runt;
155 unsigned long long bw, iops;
157 char *io_p, *bw_p, *iops_p;
160 assert(ddir_rw(ddir));
162 if (!ts->runtime[ddir])
165 i2p = is_power_of_2(rs->kb_base);
166 runt = ts->runtime[ddir];
168 bw = (1000 * ts->io_bytes[ddir]) / runt;
169 io_p = num2str(ts->io_bytes[ddir], 6, 1, i2p);
170 bw_p = num2str(bw, 6, 1, i2p);
172 iops = (1000 * (uint64_t)ts->total_io_u[ddir]) / runt;
173 iops_p = num2str(iops, 6, 1, 0);
175 log_info(" %s: io=%sB, bw=%sB/s, iops=%s, runt=%6llumsec\n",
176 ddir_str[ddir], io_p, bw_p, iops_p,
183 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) {
184 const char *base = "(usec)";
187 if (!usec_to_msec(&min, &max, &mean, &dev))
190 minp = num2str(min, 6, 1, 0);
191 maxp = num2str(max, 6, 1, 0);
193 log_info(" slat %s: min=%s, max=%s, avg=%5.02f,"
194 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
199 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) {
200 const char *base = "(usec)";
203 if (!usec_to_msec(&min, &max, &mean, &dev))
206 minp = num2str(min, 6, 1, 0);
207 maxp = num2str(max, 6, 1, 0);
209 log_info(" clat %s: min=%s, max=%s, avg=%5.02f,"
210 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
215 if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev)) {
216 const char *base = "(usec)";
219 if (!usec_to_msec(&min, &max, &mean, &dev))
222 minp = num2str(min, 6, 1, 0);
223 maxp = num2str(max, 6, 1, 0);
225 log_info(" lat %s: min=%s, max=%s, avg=%5.02f,"
226 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
231 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
234 p_of_agg = mean * 100 / (double) rs->agg[ddir];
235 log_info(" bw (KB/s) : min=%5lu, max=%5lu, per=%3.2f%%,"
236 " avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg,
241 static void show_lat(double *io_u_lat, int nr, const char **ranges,
244 int new_line = 1, i, line = 0;
246 for (i = 0; i < nr; i++) {
247 if (io_u_lat[i] <= 0.0)
252 log_info(" lat (%s): ", msg);
258 log_info("%s%3.2f%%", ranges[i], io_u_lat[i]);
265 static void show_lat_u(double *io_u_lat_u)
267 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
268 "250=", "500=", "750=", "1000=", };
270 show_lat(io_u_lat_u, FIO_IO_U_LAT_U_NR, ranges, "usec");
273 static void show_lat_m(double *io_u_lat_m)
275 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
276 "250=", "500=", "750=", "1000=", "2000=",
279 show_lat(io_u_lat_m, FIO_IO_U_LAT_M_NR, ranges, "msec");
282 static void show_latencies(double *io_u_lat_u, double *io_u_lat_m)
284 show_lat_u(io_u_lat_u);
286 show_lat_m(io_u_lat_m);
290 static void show_thread_status(struct thread_stat *ts,
291 struct group_run_stats *rs)
293 double usr_cpu, sys_cpu;
294 unsigned long runtime;
295 double io_u_dist[FIO_IO_U_MAP_NR];
296 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
297 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
299 if (!(ts->io_bytes[0] + ts->io_bytes[1]) &&
300 !(ts->total_io_u[0] + ts->total_io_u[1]))
304 log_info("%s: (groupid=%d, jobs=%d): err=%2d: pid=%d\n",
305 ts->name, ts->groupid, ts->members,
306 ts->error, (int) ts->pid);
308 log_info("%s: (groupid=%d, jobs=%d): err=%2d (%s): pid=%d\n",
309 ts->name, ts->groupid, ts->members,
310 ts->error, ts->verror, (int) ts->pid);
314 log_info(" Description : [%s]\n", ts->description);
316 if (ts->io_bytes[DDIR_READ])
317 show_ddir_status(rs, ts, DDIR_READ);
318 if (ts->io_bytes[DDIR_WRITE])
319 show_ddir_status(rs, ts, DDIR_WRITE);
321 runtime = ts->total_run_time;
323 double runt = (double) runtime;
325 usr_cpu = (double) ts->usr_time * 100 / runt;
326 sys_cpu = (double) ts->sys_time * 100 / runt;
332 log_info(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu, majf=%lu,"
333 " minf=%lu\n", usr_cpu, sys_cpu, ts->ctx, ts->majf, ts->minf);
335 stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
336 log_info(" IO depths : 1=%3.1f%%, 2=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%,"
337 " 16=%3.1f%%, 32=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
338 io_u_dist[1], io_u_dist[2],
339 io_u_dist[3], io_u_dist[4],
340 io_u_dist[5], io_u_dist[6]);
342 stat_calc_dist(ts->io_u_submit, ts->total_submit, io_u_dist);
343 log_info(" submit : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
344 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
345 io_u_dist[1], io_u_dist[2],
346 io_u_dist[3], io_u_dist[4],
347 io_u_dist[5], io_u_dist[6]);
348 stat_calc_dist(ts->io_u_complete, ts->total_complete, io_u_dist);
349 log_info(" complete : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
350 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
351 io_u_dist[1], io_u_dist[2],
352 io_u_dist[3], io_u_dist[4],
353 io_u_dist[5], io_u_dist[6]);
354 log_info(" issued r/w/d: total=%lu/%lu/%lu, short=%lu/%lu/%lu\n",
355 ts->total_io_u[0], ts->total_io_u[1],
357 ts->short_io_u[0], ts->short_io_u[1],
359 stat_calc_lat_u(ts, io_u_lat_u);
360 stat_calc_lat_m(ts, io_u_lat_m);
361 show_latencies(io_u_lat_u, io_u_lat_m);
362 if (ts->continue_on_error) {
363 log_info(" errors : total=%lu, first_error=%d/<%s>\n",
366 strerror(ts->first_error));
370 static void show_ddir_status_terse(struct thread_stat *ts,
371 struct group_run_stats *rs, int ddir)
373 unsigned long min, max;
374 unsigned long long bw;
377 assert(ddir_rw(ddir));
380 if (ts->runtime[ddir])
381 bw = ts->io_bytes[ddir] / ts->runtime[ddir];
383 log_info(";%llu;%llu;%llu", ts->io_bytes[ddir] >> 10, bw,
386 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
387 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
389 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
391 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
392 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
394 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
396 if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev))
397 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
399 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
401 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
404 p_of_agg = mean * 100 / (double) rs->agg[ddir];
405 log_info(";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
407 log_info(";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0);
410 #define FIO_TERSE_VERSION "2"
412 static void show_thread_status_terse(struct thread_stat *ts,
413 struct group_run_stats *rs)
415 double io_u_dist[FIO_IO_U_MAP_NR];
416 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
417 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
418 double usr_cpu, sys_cpu;
422 log_info("%s;%s;%d;%d", FIO_TERSE_VERSION, ts->name, ts->groupid,
424 /* Log Read Status */
425 show_ddir_status_terse(ts, rs, 0);
426 /* Log Write Status */
427 show_ddir_status_terse(ts, rs, 1);
430 if (ts->total_run_time) {
431 double runt = (double) ts->total_run_time;
433 usr_cpu = (double) ts->usr_time * 100 / runt;
434 sys_cpu = (double) ts->sys_time * 100 / runt;
440 log_info(";%f%%;%f%%;%lu;%lu;%lu", usr_cpu, sys_cpu, ts->ctx, ts->majf,
443 /* Calc % distribution of IO depths, usecond, msecond latency */
444 stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
445 stat_calc_lat_u(ts, io_u_lat_u);
446 stat_calc_lat_m(ts, io_u_lat_m);
448 /* Only show fixed 7 I/O depth levels*/
449 log_info(";%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%",
450 io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3],
451 io_u_dist[4], io_u_dist[5], io_u_dist[6]);
453 /* Microsecond latency */
454 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
455 log_info(";%3.2f%%", io_u_lat_u[i]);
456 /* Millisecond latency */
457 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
458 log_info(";%3.2f%%", io_u_lat_m[i]);
459 /* Additional output if continue_on_error set - default off*/
460 if (ts->continue_on_error)
461 log_info(";%lu;%d", ts->total_err_count, ts->first_error);
464 /* Additional output if description is set */
466 log_info(";%s", ts->description);
471 static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr)
475 dst->min_val = min(dst->min_val, src->min_val);
476 dst->max_val = max(dst->max_val, src->max_val);
479 * Compute new mean and S after the merge
480 * <http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
481 * #Parallel_algorithm>
487 double delta = src->mean - dst->mean;
489 mean = ((src->mean * src->samples) +
490 (dst->mean * dst->samples)) /
491 (dst->samples + src->samples);
493 S = src->S + dst->S + pow(delta, 2.0) *
494 (dst->samples * src->samples) /
495 (dst->samples + src->samples);
498 dst->samples += src->samples;
503 void show_run_stats(void)
505 struct group_run_stats *runstats, *rs;
506 struct thread_data *td;
507 struct thread_stat *threadstats, *ts;
508 int i, j, k, l, nr_ts, last_ts, idx;
509 int kb_base_warned = 0;
511 runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
513 for (i = 0; i < groupid + 1; i++) {
516 memset(rs, 0, sizeof(*rs));
517 rs->min_bw[0] = rs->min_run[0] = ~0UL;
518 rs->min_bw[1] = rs->min_run[1] = ~0UL;
522 * find out how many threads stats we need. if group reporting isn't
523 * enabled, it's one-per-td.
528 if (!td->o.group_reporting) {
532 if (last_ts == td->groupid)
535 last_ts = td->groupid;
539 threadstats = malloc(nr_ts * sizeof(struct thread_stat));
541 for (i = 0; i < nr_ts; i++) {
542 ts = &threadstats[i];
544 memset(ts, 0, sizeof(*ts));
545 for (j = 0; j <= DDIR_WRITE; j++) {
546 ts->lat_stat[j].min_val = -1UL;
547 ts->clat_stat[j].min_val = -1UL;
548 ts->slat_stat[j].min_val = -1UL;
549 ts->bw_stat[j].min_val = -1UL;
558 if (idx && (!td->o.group_reporting ||
559 (td->o.group_reporting && last_ts != td->groupid))) {
564 last_ts = td->groupid;
566 ts = &threadstats[j];
571 if (ts->groupid == -1) {
573 * These are per-group shared already
575 ts->name = td->o.name;
576 ts->description = td->o.description;
577 ts->groupid = td->groupid;
580 * first pid in group, not very useful...
584 ts->kb_base = td->o.kb_base;
585 } else if (ts->kb_base != td->o.kb_base && !kb_base_warned) {
586 log_info("fio: kb_base differs for jobs in group, using"
587 " %u as the base\n", ts->kb_base);
591 ts->continue_on_error = td->o.continue_on_error;
592 ts->total_err_count += td->total_err_count;
593 ts->first_error = td->first_error;
595 if (!td->error && td->o.continue_on_error &&
597 ts->error = td->first_error;
598 ts->verror = td->verror;
599 } else if (td->error) {
600 ts->error = td->error;
601 ts->verror = td->verror;
605 for (l = 0; l <= DDIR_WRITE; l++) {
606 sum_stat(&ts->clat_stat[l], &td->ts.clat_stat[l], idx);
607 sum_stat(&ts->slat_stat[l], &td->ts.slat_stat[l], idx);
608 sum_stat(&ts->lat_stat[l], &td->ts.lat_stat[l], idx);
609 sum_stat(&ts->bw_stat[l], &td->ts.bw_stat[l], idx);
611 ts->stat_io_bytes[l] += td->ts.stat_io_bytes[l];
612 ts->io_bytes[l] += td->ts.io_bytes[l];
614 if (ts->runtime[l] < td->ts.runtime[l])
615 ts->runtime[l] = td->ts.runtime[l];
618 ts->usr_time += td->ts.usr_time;
619 ts->sys_time += td->ts.sys_time;
620 ts->ctx += td->ts.ctx;
621 ts->majf += td->ts.majf;
622 ts->minf += td->ts.minf;
624 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
625 ts->io_u_map[k] += td->ts.io_u_map[k];
626 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
627 ts->io_u_submit[k] += td->ts.io_u_submit[k];
628 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
629 ts->io_u_complete[k] += td->ts.io_u_complete[k];
630 for (k = 0; k < FIO_IO_U_LAT_U_NR; k++)
631 ts->io_u_lat_u[k] += td->ts.io_u_lat_u[k];
632 for (k = 0; k < FIO_IO_U_LAT_M_NR; k++)
633 ts->io_u_lat_m[k] += td->ts.io_u_lat_m[k];
636 for (k = 0; k <= 2; k++) {
637 ts->total_io_u[k] += td->ts.total_io_u[k];
638 ts->short_io_u[k] += td->ts.short_io_u[k];
641 ts->total_run_time += td->ts.total_run_time;
642 ts->total_submit += td->ts.total_submit;
643 ts->total_complete += td->ts.total_complete;
646 for (i = 0; i < nr_ts; i++) {
647 unsigned long long bw;
649 ts = &threadstats[i];
650 rs = &runstats[ts->groupid];
651 rs->kb_base = ts->kb_base;
653 for (j = 0; j <= DDIR_WRITE; j++) {
656 if (ts->runtime[j] < rs->min_run[j] || !rs->min_run[j])
657 rs->min_run[j] = ts->runtime[j];
658 if (ts->runtime[j] > rs->max_run[j])
659 rs->max_run[j] = ts->runtime[j];
662 if (ts->runtime[j]) {
665 runt = ts->runtime[j];
666 bw = ts->io_bytes[j] / runt;
668 if (bw < rs->min_bw[j])
670 if (bw > rs->max_bw[j])
673 rs->io_kb[j] += ts->io_bytes[j] / rs->kb_base;
677 for (i = 0; i < groupid + 1; i++) {
678 unsigned long max_run[2];
681 max_run[0] = rs->max_run[0];
682 max_run[1] = rs->max_run[1];
685 rs->agg[0] = (rs->io_kb[0] * 1000) / max_run[0];
687 rs->agg[1] = (rs->io_kb[1] * 1000) / max_run[1];
691 * don't overwrite last signal output
696 for (i = 0; i < nr_ts; i++) {
697 ts = &threadstats[i];
698 rs = &runstats[ts->groupid];
701 show_thread_status_terse(ts, rs);
703 show_thread_status(ts, rs);
707 for (i = 0; i < groupid + 1; i++)
708 show_group_stats(&runstats[i], i);
717 static inline void add_stat_sample(struct io_stat *is, unsigned long data)
722 if (data > is->max_val)
724 if (data < is->min_val)
727 delta = val - is->mean;
729 is->mean += delta / (is->samples + 1.0);
730 is->S += delta * (val - is->mean);
736 static void __add_log_sample(struct io_log *iolog, unsigned long val,
737 enum fio_ddir ddir, unsigned int bs,
740 const int nr_samples = iolog->nr_samples;
742 if (iolog->nr_samples == iolog->max_samples) {
743 int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
745 iolog->log = realloc(iolog->log, new_size);
746 iolog->max_samples <<= 1;
749 iolog->log[nr_samples].val = val;
750 iolog->log[nr_samples].time = t;
751 iolog->log[nr_samples].ddir = ddir;
752 iolog->log[nr_samples].bs = bs;
756 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
757 unsigned long val, enum fio_ddir ddir,
763 __add_log_sample(iolog, val, ddir, bs, mtime_since_now(&td->epoch));
766 void add_agg_sample(unsigned long val, enum fio_ddir ddir, unsigned int bs)
768 struct io_log *iolog;
773 iolog = agg_io_log[ddir];
774 __add_log_sample(iolog, val, ddir, bs, mtime_since_genesis());
777 void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
778 unsigned long usec, unsigned int bs)
780 struct thread_stat *ts = &td->ts;
785 add_stat_sample(&ts->clat_stat[ddir], usec);
788 add_log_sample(td, ts->clat_log, usec, ddir, bs);
791 void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
792 unsigned long usec, unsigned int bs)
794 struct thread_stat *ts = &td->ts;
799 add_stat_sample(&ts->slat_stat[ddir], usec);
802 add_log_sample(td, ts->slat_log, usec, ddir, bs);
805 void add_lat_sample(struct thread_data *td, enum fio_ddir ddir,
806 unsigned long usec, unsigned int bs)
808 struct thread_stat *ts = &td->ts;
813 add_stat_sample(&ts->lat_stat[ddir], usec);
816 add_log_sample(td, ts->lat_log, usec, ddir, bs);
819 void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs,
822 struct thread_stat *ts = &td->ts;
823 unsigned long spent, rate;
828 spent = mtime_since(&ts->stat_sample_time[ddir], t);
829 if (spent < td->o.bw_avg_time)
832 rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) *
834 add_stat_sample(&ts->bw_stat[ddir], rate);
837 add_log_sample(td, ts->bw_log, rate, ddir, bs);
839 fio_gettime(&ts->stat_sample_time[ddir], NULL);
840 ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir];