13 * Cheesy number->string conversion, complete with carry rounding error.
15 static char *num2str(unsigned long num, int maxlen, int base, int pow2)
17 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
18 unsigned int thousand;
29 for (i = 0; base > 1; i++)
35 len = sprintf(buf, "%'lu", num);
38 buf[len] = postfix[i];
44 if ((num % thousand) >= (thousand / 2))
55 void update_rusage_stat(struct thread_data *td)
57 struct thread_stat *ts = &td->ts;
59 getrusage(RUSAGE_SELF, &ts->ru_end);
61 ts->usr_time += mtime_since(&ts->ru_start.ru_utime, &ts->ru_end.ru_utime);
62 ts->sys_time += mtime_since(&ts->ru_start.ru_stime, &ts->ru_end.ru_stime);
63 ts->ctx += ts->ru_end.ru_nvcsw + ts->ru_end.ru_nivcsw - (ts->ru_start.ru_nvcsw + ts->ru_start.ru_nivcsw);
64 ts->minf += ts->ru_end.ru_minflt - ts->ru_start.ru_minflt;
65 ts->majf += ts->ru_end.ru_majflt - ts->ru_start.ru_majflt;
67 memcpy(&ts->ru_start, &ts->ru_end, sizeof(ts->ru_end));
70 static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
71 double *mean, double *dev)
73 double n = is->samples;
81 n = (double) is->samples;
85 *dev = sqrt(is->S / (n - 1.0));
92 static void show_group_stats(struct group_run_stats *rs, int id)
94 char *p1, *p2, *p3, *p4;
95 const char *ddir_str[] = { " READ", " WRITE" };
98 log_info("\nRun status group %d (all jobs):\n", id);
100 for (i = 0; i <= DDIR_WRITE; i++) {
104 p1 = num2str(rs->io_kb[i], 6, 1000, 1);
105 p2 = num2str(rs->agg[i], 6, 1000, 1);
106 p3 = num2str(rs->min_bw[i], 6, 1000, 1);
107 p4 = num2str(rs->max_bw[i], 6, 1000, 1);
109 log_info("%s: io=%siB, aggrb=%siB/s, minb=%siB/s, maxb=%siB/s, mint=%llumsec, maxt=%llumsec\n", ddir_str[i], p1, p2, p3, p4, rs->min_run[i], rs->max_run[i]);
118 #define ts_total_io_u(ts) \
119 ((ts)->total_io_u[0] + (ts)->total_io_u[1])
121 static void stat_calc_dist(struct thread_stat *ts, double *io_u_dist)
126 * Do depth distribution calculations
128 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
129 io_u_dist[i] = (double) ts->io_u_map[i] / (double) ts_total_io_u(ts);
130 io_u_dist[i] *= 100.0;
131 if (io_u_dist[i] < 0.1 && ts->io_u_map[i])
136 static void stat_calc_lat(struct thread_stat *ts, double *dst,
137 unsigned int *src, int nr)
142 * Do latency distribution calculations
144 for (i = 0; i < nr; i++) {
145 dst[i] = (double) src[i] / (double) ts_total_io_u(ts);
147 if (dst[i] < 0.01 && src[i])
152 static void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat)
154 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_u, FIO_IO_U_LAT_U_NR);
157 static void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat)
159 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_m, FIO_IO_U_LAT_M_NR);
162 static int usec_to_msec(unsigned long *min, unsigned long *max, double *mean,
165 if (*min > 1000 && *max > 1000 && *mean > 1000.0 && *dev > 1000.0) {
176 static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
179 const char *ddir_str[] = { "read ", "write" };
180 unsigned long min, max;
181 unsigned long long bw, iops;
183 char *io_p, *bw_p, *iops_p;
185 if (!ts->runtime[ddir])
188 bw = ts->io_bytes[ddir] / ts->runtime[ddir];
189 iops = (1000 * ts->total_io_u[ddir]) / ts->runtime[ddir];
190 io_p = num2str(ts->io_bytes[ddir] >> 10, 6, 1000, 1);
191 bw_p = num2str(bw, 6, 1000, 1);
192 iops_p = num2str(iops, 6, 1, 0);
194 log_info(" %s: io=%siB, bw=%siB/s, iops=%s, runt=%6lumsec\n", ddir_str[ddir], io_p, bw_p, iops_p, ts->runtime[ddir]);
200 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) {
201 const char *base = "(usec)";
204 if (!usec_to_msec(&min, &max, &mean, &dev))
207 minp = num2str(min, 6, 1, 0);
208 maxp = num2str(max, 6, 1, 0);
210 log_info(" slat %s: min=%s, max=%s, avg=%5.02f, stdev=%5.02f\n", base, minp, maxp, mean, dev);
215 if (calc_lat(&ts->clat_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(" clat %s: min=%s, max=%s, avg=%5.02f, stdev=%5.02f\n", base, minp, maxp, mean, dev);
230 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
233 p_of_agg = mean * 100 / (double) rs->agg[ddir];
234 log_info(" bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg, mean, dev);
238 static void show_lat(double *io_u_lat, int nr, const char **ranges,
241 int new_line = 1, i, line = 0;
243 for (i = 0; i < nr; i++) {
244 if (io_u_lat[i] <= 0.0)
249 log_info(" lat (%s): ", msg);
255 log_info("%s%3.2f%%", ranges[i], io_u_lat[i]);
262 static void show_lat_u(double *io_u_lat_u)
264 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
265 "250=", "500=", "750=", "1000=", };
267 show_lat(io_u_lat_u, FIO_IO_U_LAT_U_NR, ranges, "usec");
270 static void show_lat_m(double *io_u_lat_m)
272 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
273 "250=", "500=", "750=", "1000=", "2000=",
276 show_lat(io_u_lat_m, FIO_IO_U_LAT_M_NR, ranges, "msec");
279 static void show_latencies(double *io_u_lat_u, double *io_u_lat_m)
281 show_lat_u(io_u_lat_u);
283 show_lat_m(io_u_lat_m);
287 static void show_thread_status(struct thread_stat *ts,
288 struct group_run_stats *rs)
290 double usr_cpu, sys_cpu;
291 unsigned long runtime;
292 double io_u_dist[FIO_IO_U_MAP_NR];
293 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
294 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
296 if (!(ts->io_bytes[0] + ts->io_bytes[1]) &&
297 !(ts->total_io_u[0] + ts->total_io_u[1]))
301 log_info("%s: (groupid=%d, jobs=%d): err=%2d: pid=%d\n", ts->name, ts->groupid, ts->members, ts->error, ts->pid);
303 log_info("%s: (groupid=%d, jobs=%d): err=%2d (%s): pid=%d\n", ts->name, ts->groupid, ts->members, ts->error, ts->verror, ts->pid);
306 log_info(" Description : [%s]\n", ts->description);
308 if (ts->io_bytes[DDIR_READ])
309 show_ddir_status(rs, ts, DDIR_READ);
310 if (ts->io_bytes[DDIR_WRITE])
311 show_ddir_status(rs, ts, DDIR_WRITE);
313 runtime = ts->total_run_time;
315 double runt = (double) runtime;
317 usr_cpu = (double) ts->usr_time * 100 / runt;
318 sys_cpu = (double) ts->sys_time * 100 / runt;
324 log_info(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu, majf=%lu, minf=%lu\n", usr_cpu, sys_cpu, ts->ctx, ts->majf, ts->minf);
326 stat_calc_dist(ts, io_u_dist);
327 stat_calc_lat_u(ts, io_u_lat_u);
328 stat_calc_lat_m(ts, io_u_lat_m);
330 log_info(" IO depths : 1=%3.1f%%, 2=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%, 32=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3], io_u_dist[4], io_u_dist[5], io_u_dist[6]);
331 log_info(" issued r/w: total=%lu/%lu, short=%lu/%lu\n", ts->total_io_u[0], ts->total_io_u[1], ts->short_io_u[0], ts->short_io_u[1]);
333 show_latencies(io_u_lat_u, io_u_lat_m);
336 static void show_ddir_status_terse(struct thread_stat *ts,
337 struct group_run_stats *rs, int ddir)
339 unsigned long min, max;
340 unsigned long long bw;
344 if (ts->runtime[ddir])
345 bw = ts->io_bytes[ddir] / ts->runtime[ddir];
347 log_info(";%llu;%llu;%lu", ts->io_bytes[ddir] >> 10, bw, ts->runtime[ddir]);
349 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
350 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
352 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
354 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
355 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
357 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
359 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
362 p_of_agg = mean * 100 / (double) rs->agg[ddir];
363 log_info(";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
365 log_info(";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0);
369 static void show_thread_status_terse(struct thread_stat *ts,
370 struct group_run_stats *rs)
372 double io_u_dist[FIO_IO_U_MAP_NR];
373 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
374 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
375 double usr_cpu, sys_cpu;
378 log_info("%s;%d;%d", ts->name, ts->groupid, ts->error);
380 show_ddir_status_terse(ts, rs, 0);
381 show_ddir_status_terse(ts, rs, 1);
383 if (ts->total_run_time) {
384 double runt = (double) ts->total_run_time;
386 usr_cpu = (double) ts->usr_time * 100 / runt;
387 sys_cpu = (double) ts->sys_time * 100 / runt;
393 log_info(";%f%%;%f%%;%lu;%lu;%lu", usr_cpu, sys_cpu, ts->ctx, ts->majf, ts->minf);
395 stat_calc_dist(ts, io_u_dist);
396 stat_calc_lat_u(ts, io_u_lat_u);
397 stat_calc_lat_m(ts, io_u_lat_m);
399 log_info(";%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%", io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3], io_u_dist[4], io_u_dist[5], io_u_dist[6]);
401 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
402 log_info(";%3.2f%%", io_u_lat_u[i]);
403 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
404 log_info(";%3.2f%%", io_u_lat_m[i]);
408 log_info(";%s", ts->description);
413 static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr)
417 dst->min_val = min(dst->min_val, src->min_val);
418 dst->max_val = max(dst->max_val, src->max_val);
419 dst->samples += src->samples;
422 * Needs a new method for calculating stddev, we cannot just
423 * average them we do below for nr > 1
429 mean = ((src->mean * (double) (nr - 1)) + dst->mean) / ((double) nr);
430 S = ((src->S * (double) (nr - 1)) + dst->S) / ((double) nr);
437 void show_run_stats(void)
439 struct group_run_stats *runstats, *rs;
440 struct thread_data *td;
441 struct thread_stat *threadstats, *ts;
442 int i, j, k, l, nr_ts, last_ts, idx;
444 runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
446 for (i = 0; i < groupid + 1; i++) {
449 memset(rs, 0, sizeof(*rs));
450 rs->min_bw[0] = rs->min_run[0] = ~0UL;
451 rs->min_bw[1] = rs->min_run[1] = ~0UL;
455 * find out how many threads stats we need. if group reporting isn't
456 * enabled, it's one-per-td.
461 if (!td->o.group_reporting) {
465 if (last_ts == td->groupid)
468 last_ts = td->groupid;
472 threadstats = malloc(nr_ts * sizeof(struct thread_stat));
474 for (i = 0; i < nr_ts; i++) {
475 ts = &threadstats[i];
477 memset(ts, 0, sizeof(*ts));
478 for (j = 0; j <= DDIR_WRITE; j++) {
479 ts->clat_stat[j].min_val = -1UL;
480 ts->slat_stat[j].min_val = -1UL;
481 ts->bw_stat[j].min_val = -1UL;
490 if (idx && (!td->o.group_reporting ||
491 (td->o.group_reporting && last_ts != td->groupid))) {
496 last_ts = td->groupid;
498 ts = &threadstats[j];
503 if (ts->groupid == -1) {
505 * These are per-group shared already
507 ts->name = td->o.name;
508 ts->description = td->o.description;
509 ts->groupid = td->groupid;
512 * first pid in group, not very useful...
517 if (td->error && !ts->error) {
518 ts->error = td->error;
519 ts->verror = td->verror;
522 for (l = 0; l <= DDIR_WRITE; l++) {
523 sum_stat(&ts->clat_stat[l], &td->ts.clat_stat[l], idx);
524 sum_stat(&ts->slat_stat[l], &td->ts.slat_stat[l], idx);
525 sum_stat(&ts->bw_stat[l], &td->ts.bw_stat[l], idx);
527 ts->stat_io_bytes[l] += td->ts.stat_io_bytes[l];
528 ts->io_bytes[l] += td->ts.io_bytes[l];
530 if (ts->runtime[l] < td->ts.runtime[l])
531 ts->runtime[l] = td->ts.runtime[l];
534 ts->usr_time += td->ts.usr_time;
535 ts->sys_time += td->ts.sys_time;
536 ts->ctx += td->ts.ctx;
537 ts->majf += td->ts.majf;
538 ts->minf += td->ts.minf;
540 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
541 ts->io_u_map[k] += td->ts.io_u_map[k];
542 for (k = 0; k < FIO_IO_U_LAT_U_NR; k++)
543 ts->io_u_lat_u[k] += td->ts.io_u_lat_u[k];
544 for (k = 0; k < FIO_IO_U_LAT_M_NR; k++)
545 ts->io_u_lat_m[k] += td->ts.io_u_lat_m[k];
548 for (k = 0; k <= DDIR_WRITE; k++) {
549 ts->total_io_u[k] += td->ts.total_io_u[k];
550 ts->short_io_u[k] += td->ts.short_io_u[k];
553 ts->total_run_time += td->ts.total_run_time;
556 for (i = 0; i < nr_ts; i++) {
557 unsigned long long bw;
559 ts = &threadstats[i];
560 rs = &runstats[ts->groupid];
562 for (j = 0; j <= DDIR_WRITE; j++) {
565 if (ts->runtime[j] < rs->min_run[j] || !rs->min_run[j])
566 rs->min_run[j] = ts->runtime[j];
567 if (ts->runtime[j] > rs->max_run[j])
568 rs->max_run[j] = ts->runtime[j];
572 bw = ts->io_bytes[j] / (unsigned long long) ts->runtime[j];
573 if (bw < rs->min_bw[j])
575 if (bw > rs->max_bw[j])
578 rs->io_kb[j] += ts->io_bytes[j] >> 10;
582 for (i = 0; i < groupid + 1; i++) {
586 rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0];
588 rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1];
592 * don't overwrite last signal output
597 for (i = 0; i < nr_ts; i++) {
598 ts = &threadstats[i];
599 rs = &runstats[ts->groupid];
602 show_thread_status_terse(ts, rs);
604 show_thread_status(ts, rs);
608 for (i = 0; i < groupid + 1; i++)
609 show_group_stats(&runstats[i], i);
618 static inline void add_stat_sample(struct io_stat *is, unsigned long data)
623 if (data > is->max_val)
625 if (data < is->min_val)
628 delta = val - is->mean;
630 is->mean += delta / (is->samples + 1.0);
631 is->S += delta * (val - is->mean);
637 static void __add_log_sample(struct io_log *iolog, unsigned long val,
638 enum fio_ddir ddir, unsigned long time)
640 if (iolog->nr_samples == iolog->max_samples) {
641 int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
643 iolog->log = realloc(iolog->log, new_size);
644 iolog->max_samples <<= 1;
647 iolog->log[iolog->nr_samples].val = val;
648 iolog->log[iolog->nr_samples].time = time;
649 iolog->log[iolog->nr_samples].ddir = ddir;
653 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
654 unsigned long val, enum fio_ddir ddir)
656 __add_log_sample(iolog, val, ddir, mtime_since_now(&td->epoch));
659 void add_agg_sample(unsigned long val, enum fio_ddir ddir)
661 struct io_log *iolog = agg_io_log[ddir];
663 __add_log_sample(iolog, val, ddir, mtime_since_genesis());
666 void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
669 struct thread_stat *ts = &td->ts;
671 add_stat_sample(&ts->clat_stat[ddir], usec);
674 add_log_sample(td, ts->clat_log, usec, ddir);
677 void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
680 struct thread_stat *ts = &td->ts;
682 add_stat_sample(&ts->slat_stat[ddir], usec);
685 add_log_sample(td, ts->slat_log, usec, ddir);
688 void add_bw_sample(struct thread_data *td, enum fio_ddir ddir,
691 struct thread_stat *ts = &td->ts;
692 unsigned long spent = mtime_since(&ts->stat_sample_time[ddir], t);
695 if (spent < td->o.bw_avg_time)
698 rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) / spent;
699 add_stat_sample(&ts->bw_stat[ddir], rate);
702 add_log_sample(td, ts->bw_log, rate, ddir);
704 fio_gettime(&ts->stat_sample_time[ddir], NULL);
705 ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir];