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 if (!ts->runtime[ddir])
163 i2p = is_power_of_2(rs->kb_base);
164 runt = ts->runtime[ddir];
166 bw = (1000 * ts->io_bytes[ddir]) / runt;
167 io_p = num2str(ts->io_bytes[ddir], 6, 1, i2p);
168 bw_p = num2str(bw, 6, 1, i2p);
170 iops = (1000 * ts->total_io_u[ddir]) / runt;
171 iops_p = num2str(iops, 6, 1, 0);
173 log_info(" %s: io=%sB, bw=%sB/s, iops=%s, runt=%6lumsec\n",
174 ddir_str[ddir], io_p, bw_p, iops_p,
181 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) {
182 const char *base = "(usec)";
185 if (!usec_to_msec(&min, &max, &mean, &dev))
188 minp = num2str(min, 6, 1, 0);
189 maxp = num2str(max, 6, 1, 0);
191 log_info(" slat %s: min=%s, max=%s, avg=%5.02f,"
192 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
197 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) {
198 const char *base = "(usec)";
201 if (!usec_to_msec(&min, &max, &mean, &dev))
204 minp = num2str(min, 6, 1, 0);
205 maxp = num2str(max, 6, 1, 0);
207 log_info(" clat %s: min=%s, max=%s, avg=%5.02f,"
208 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
213 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
216 p_of_agg = mean * 100 / (double) rs->agg[ddir];
217 log_info(" bw (KB/s) : min=%5lu, max=%5lu, per=%3.2f%%,"
218 " avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg,
223 static void show_lat(double *io_u_lat, int nr, const char **ranges,
226 int new_line = 1, i, line = 0;
228 for (i = 0; i < nr; i++) {
229 if (io_u_lat[i] <= 0.0)
234 log_info(" lat (%s): ", msg);
240 log_info("%s%3.2f%%", ranges[i], io_u_lat[i]);
247 static void show_lat_u(double *io_u_lat_u)
249 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
250 "250=", "500=", "750=", "1000=", };
252 show_lat(io_u_lat_u, FIO_IO_U_LAT_U_NR, ranges, "usec");
255 static void show_lat_m(double *io_u_lat_m)
257 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
258 "250=", "500=", "750=", "1000=", "2000=",
261 show_lat(io_u_lat_m, FIO_IO_U_LAT_M_NR, ranges, "msec");
264 static void show_latencies(double *io_u_lat_u, double *io_u_lat_m)
266 show_lat_u(io_u_lat_u);
268 show_lat_m(io_u_lat_m);
272 static void show_thread_status(struct thread_stat *ts,
273 struct group_run_stats *rs)
275 double usr_cpu, sys_cpu;
276 unsigned long runtime;
277 double io_u_dist[FIO_IO_U_MAP_NR];
278 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
279 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
281 if (!(ts->io_bytes[0] + ts->io_bytes[1]) &&
282 !(ts->total_io_u[0] + ts->total_io_u[1]))
286 log_info("%s: (groupid=%d, jobs=%d): err=%2d: pid=%d\n",
287 ts->name, ts->groupid, ts->members,
288 ts->error, (int) ts->pid);
290 log_info("%s: (groupid=%d, jobs=%d): err=%2d (%s): pid=%d\n",
291 ts->name, ts->groupid, ts->members,
292 ts->error, ts->verror, (int) ts->pid);
296 log_info(" Description : [%s]\n", ts->description);
298 if (ts->io_bytes[DDIR_READ])
299 show_ddir_status(rs, ts, DDIR_READ);
300 if (ts->io_bytes[DDIR_WRITE])
301 show_ddir_status(rs, ts, DDIR_WRITE);
303 runtime = ts->total_run_time;
305 double runt = (double) runtime;
307 usr_cpu = (double) ts->usr_time * 100 / runt;
308 sys_cpu = (double) ts->sys_time * 100 / runt;
314 log_info(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu, majf=%lu,"
315 " minf=%lu\n", usr_cpu, sys_cpu, ts->ctx, ts->majf, ts->minf);
317 stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
318 log_info(" IO depths : 1=%3.1f%%, 2=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%,"
319 " 16=%3.1f%%, 32=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
320 io_u_dist[1], io_u_dist[2],
321 io_u_dist[3], io_u_dist[4],
322 io_u_dist[5], io_u_dist[6]);
324 stat_calc_dist(ts->io_u_submit, ts->total_submit, io_u_dist);
325 log_info(" submit : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
326 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
327 io_u_dist[1], io_u_dist[2],
328 io_u_dist[3], io_u_dist[4],
329 io_u_dist[5], io_u_dist[6]);
330 stat_calc_dist(ts->io_u_complete, ts->total_complete, io_u_dist);
331 log_info(" complete : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
332 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
333 io_u_dist[1], io_u_dist[2],
334 io_u_dist[3], io_u_dist[4],
335 io_u_dist[5], io_u_dist[6]);
336 log_info(" issued r/w: total=%lu/%lu, short=%lu/%lu\n",
337 ts->total_io_u[0], ts->total_io_u[1],
338 ts->short_io_u[0], ts->short_io_u[1]);
339 stat_calc_lat_u(ts, io_u_lat_u);
340 stat_calc_lat_m(ts, io_u_lat_m);
341 show_latencies(io_u_lat_u, io_u_lat_m);
342 if (ts->continue_on_error) {
343 log_info(" errors : total=%lu, first_error=%d/<%s>\n",
346 strerror(ts->first_error));
350 static void show_ddir_status_terse(struct thread_stat *ts,
351 struct group_run_stats *rs, int ddir)
353 unsigned long min, max;
354 unsigned long long bw;
358 if (ts->runtime[ddir])
359 bw = ts->io_bytes[ddir] / ts->runtime[ddir];
361 log_info(";%llu;%llu;%lu", ts->io_bytes[ddir] >> 10, bw,
364 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
365 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
367 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
369 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
370 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
372 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
374 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
377 p_of_agg = mean * 100 / (double) rs->agg[ddir];
378 log_info(";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
380 log_info(";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0);
384 static void show_thread_status_terse(struct thread_stat *ts,
385 struct group_run_stats *rs)
387 double io_u_dist[FIO_IO_U_MAP_NR];
388 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
389 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
390 double usr_cpu, sys_cpu;
393 log_info("%s;%d;%d", ts->name, ts->groupid, ts->error);
395 show_ddir_status_terse(ts, rs, 0);
396 show_ddir_status_terse(ts, rs, 1);
398 if (ts->total_run_time) {
399 double runt = (double) ts->total_run_time;
401 usr_cpu = (double) ts->usr_time * 100 / runt;
402 sys_cpu = (double) ts->sys_time * 100 / runt;
408 log_info(";%f%%;%f%%;%lu;%lu;%lu", usr_cpu, sys_cpu, ts->ctx, ts->majf,
411 stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
412 stat_calc_lat_u(ts, io_u_lat_u);
413 stat_calc_lat_m(ts, io_u_lat_m);
415 log_info(";%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%",
416 io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3],
417 io_u_dist[4], io_u_dist[5], io_u_dist[6]);
419 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
420 log_info(";%3.2f%%", io_u_lat_u[i]);
421 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
422 log_info(";%3.2f%%", io_u_lat_m[i]);
423 if (ts->continue_on_error)
424 log_info(";%lu;%d", ts->total_err_count, ts->first_error);
428 log_info(";%s", ts->description);
433 static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr)
437 dst->min_val = min(dst->min_val, src->min_val);
438 dst->max_val = max(dst->max_val, src->max_val);
439 dst->samples += src->samples;
442 * Needs a new method for calculating stddev, we cannot just
443 * average them we do below for nr > 1
449 mean = ((src->mean * (double) (nr - 1))
450 + dst->mean) / ((double) nr);
451 S = ((src->S * (double) (nr - 1)) + dst->S) / ((double) nr);
458 void show_run_stats(void)
460 struct group_run_stats *runstats, *rs;
461 struct thread_data *td;
462 struct thread_stat *threadstats, *ts;
463 int i, j, k, l, nr_ts, last_ts, idx;
464 int kb_base_warned = 0;
466 runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
468 for (i = 0; i < groupid + 1; i++) {
471 memset(rs, 0, sizeof(*rs));
472 rs->min_bw[0] = rs->min_run[0] = ~0UL;
473 rs->min_bw[1] = rs->min_run[1] = ~0UL;
477 * find out how many threads stats we need. if group reporting isn't
478 * enabled, it's one-per-td.
483 if (!td->o.group_reporting) {
487 if (last_ts == td->groupid)
490 last_ts = td->groupid;
494 threadstats = malloc(nr_ts * sizeof(struct thread_stat));
496 for (i = 0; i < nr_ts; i++) {
497 ts = &threadstats[i];
499 memset(ts, 0, sizeof(*ts));
500 for (j = 0; j <= DDIR_WRITE; j++) {
501 ts->clat_stat[j].min_val = -1UL;
502 ts->slat_stat[j].min_val = -1UL;
503 ts->bw_stat[j].min_val = -1UL;
512 if (idx && (!td->o.group_reporting ||
513 (td->o.group_reporting && last_ts != td->groupid))) {
518 last_ts = td->groupid;
520 ts = &threadstats[j];
525 if (ts->groupid == -1) {
527 * These are per-group shared already
529 ts->name = td->o.name;
530 ts->description = td->o.description;
531 ts->groupid = td->groupid;
534 * first pid in group, not very useful...
538 ts->kb_base = td->o.kb_base;
539 } else if (ts->kb_base != td->o.kb_base && !kb_base_warned) {
540 log_info("fio: kb_base differs for jobs in group, using"
541 " %u as the base\n", ts->kb_base);
545 ts->continue_on_error = td->o.continue_on_error;
546 ts->total_err_count += td->total_err_count;
547 ts->first_error = td->first_error;
549 if (!td->error && td->o.continue_on_error &&
551 ts->error = td->first_error;
552 ts->verror = td->verror;
553 } else if (td->error) {
554 ts->error = td->error;
555 ts->verror = td->verror;
559 for (l = 0; l <= DDIR_WRITE; l++) {
560 sum_stat(&ts->clat_stat[l], &td->ts.clat_stat[l], idx);
561 sum_stat(&ts->slat_stat[l], &td->ts.slat_stat[l], idx);
562 sum_stat(&ts->bw_stat[l], &td->ts.bw_stat[l], idx);
564 ts->stat_io_bytes[l] += td->ts.stat_io_bytes[l];
565 ts->io_bytes[l] += td->ts.io_bytes[l];
567 if (ts->runtime[l] < td->ts.runtime[l])
568 ts->runtime[l] = td->ts.runtime[l];
571 ts->usr_time += td->ts.usr_time;
572 ts->sys_time += td->ts.sys_time;
573 ts->ctx += td->ts.ctx;
574 ts->majf += td->ts.majf;
575 ts->minf += td->ts.minf;
577 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
578 ts->io_u_map[k] += td->ts.io_u_map[k];
579 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
580 ts->io_u_submit[k] += td->ts.io_u_submit[k];
581 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
582 ts->io_u_complete[k] += td->ts.io_u_complete[k];
583 for (k = 0; k < FIO_IO_U_LAT_U_NR; k++)
584 ts->io_u_lat_u[k] += td->ts.io_u_lat_u[k];
585 for (k = 0; k < FIO_IO_U_LAT_M_NR; k++)
586 ts->io_u_lat_m[k] += td->ts.io_u_lat_m[k];
589 for (k = 0; k <= DDIR_WRITE; k++) {
590 ts->total_io_u[k] += td->ts.total_io_u[k];
591 ts->short_io_u[k] += td->ts.short_io_u[k];
594 ts->total_run_time += td->ts.total_run_time;
595 ts->total_submit += td->ts.total_submit;
596 ts->total_complete += td->ts.total_complete;
599 for (i = 0; i < nr_ts; i++) {
600 unsigned long long bw;
602 ts = &threadstats[i];
603 rs = &runstats[ts->groupid];
604 rs->kb_base = ts->kb_base;
606 for (j = 0; j <= DDIR_WRITE; j++) {
609 if (ts->runtime[j] < rs->min_run[j] || !rs->min_run[j])
610 rs->min_run[j] = ts->runtime[j];
611 if (ts->runtime[j] > rs->max_run[j])
612 rs->max_run[j] = ts->runtime[j];
615 if (ts->runtime[j]) {
618 runt = ts->runtime[j];
619 bw = ts->io_bytes[j] / runt;
621 if (bw < rs->min_bw[j])
623 if (bw > rs->max_bw[j])
626 rs->io_kb[j] += ts->io_bytes[j] / rs->kb_base;
630 for (i = 0; i < groupid + 1; i++) {
631 unsigned long max_run[2];
634 max_run[0] = rs->max_run[0];
635 max_run[1] = rs->max_run[1];
638 rs->agg[0] = (rs->io_kb[0] * 1000) / max_run[0];
640 rs->agg[1] = (rs->io_kb[1] * 1000) / max_run[1];
644 * don't overwrite last signal output
649 for (i = 0; i < nr_ts; i++) {
650 ts = &threadstats[i];
651 rs = &runstats[ts->groupid];
654 show_thread_status_terse(ts, rs);
656 show_thread_status(ts, rs);
660 for (i = 0; i < groupid + 1; i++)
661 show_group_stats(&runstats[i], i);
670 static inline void add_stat_sample(struct io_stat *is, unsigned long data)
675 if (data > is->max_val)
677 if (data < is->min_val)
680 delta = val - is->mean;
682 is->mean += delta / (is->samples + 1.0);
683 is->S += delta * (val - is->mean);
689 static void __add_log_sample(struct io_log *iolog, unsigned long val,
690 enum fio_ddir ddir, unsigned int bs,
693 const int nr_samples = iolog->nr_samples;
695 if (iolog->nr_samples == iolog->max_samples) {
696 int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
698 iolog->log = realloc(iolog->log, new_size);
699 iolog->max_samples <<= 1;
702 iolog->log[nr_samples].val = val;
703 iolog->log[nr_samples].time = time;
704 iolog->log[nr_samples].ddir = ddir;
705 iolog->log[nr_samples].bs = bs;
709 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
710 unsigned long val, enum fio_ddir ddir,
713 __add_log_sample(iolog, val, ddir, bs, mtime_since_now(&td->epoch));
716 void add_agg_sample(unsigned long val, enum fio_ddir ddir, unsigned int bs)
718 struct io_log *iolog = agg_io_log[ddir];
720 __add_log_sample(iolog, val, ddir, bs, mtime_since_genesis());
723 void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
724 unsigned long usec, unsigned int bs)
726 struct thread_stat *ts = &td->ts;
728 add_stat_sample(&ts->clat_stat[ddir], usec);
731 add_log_sample(td, ts->clat_log, usec, ddir, bs);
734 void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
735 unsigned long usec, unsigned int bs)
737 struct thread_stat *ts = &td->ts;
739 add_stat_sample(&ts->slat_stat[ddir], usec);
742 add_log_sample(td, ts->slat_log, usec, ddir, bs);
745 void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs,
748 struct thread_stat *ts = &td->ts;
749 unsigned long spent = mtime_since(&ts->stat_sample_time[ddir], t);
752 if (spent < td->o.bw_avg_time)
755 rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) * 1000 / spent / 1024;
756 add_stat_sample(&ts->bw_stat[ddir], rate);
759 add_log_sample(td, ts->bw_log, rate, ddir, bs);
761 fio_gettime(&ts->stat_sample_time[ddir], NULL);
762 ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir];