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);
65 memcpy(&ts->ru_start, &ts->ru_end, sizeof(ts->ru_end));
68 static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
69 double *mean, double *dev)
71 double n = is->samples;
79 n = (double) is->samples;
83 *dev = sqrt(is->S / (n - 1.0));
90 static void show_group_stats(struct group_run_stats *rs, int id)
92 char *p1, *p2, *p3, *p4;
93 const char *ddir_str[] = { " READ", " WRITE" };
96 log_info("\nRun status group %d (all jobs):\n", id);
98 for (i = 0; i <= DDIR_WRITE; i++) {
102 p1 = num2str(rs->io_kb[i], 6, 1000, 1);
103 p2 = num2str(rs->agg[i], 6, 1000, 1);
104 p3 = num2str(rs->min_bw[i], 6, 1000, 1);
105 p4 = num2str(rs->max_bw[i], 6, 1000, 1);
107 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]);
116 #define ts_total_io_u(ts) \
117 ((ts)->total_io_u[0] + (ts)->total_io_u[1])
119 static void stat_calc_dist(struct thread_stat *ts, double *io_u_dist)
124 * Do depth distribution calculations
126 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
127 io_u_dist[i] = (double) ts->io_u_map[i] / (double) ts_total_io_u(ts);
128 io_u_dist[i] *= 100.0;
129 if (io_u_dist[i] < 0.1 && ts->io_u_map[i])
134 static void stat_calc_lat(struct thread_stat *ts, double *dst,
135 unsigned int *src, int nr)
140 * Do latency distribution calculations
142 for (i = 0; i < nr; i++) {
143 dst[i] = (double) src[i] / (double) ts_total_io_u(ts);
145 if (dst[i] < 0.01 && src[i])
150 static void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat)
152 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_u, FIO_IO_U_LAT_U_NR);
155 static void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat)
157 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_m, FIO_IO_U_LAT_M_NR);
160 static int usec_to_msec(unsigned long *min, unsigned long *max, double *mean,
163 if (*min > 1000 && *max > 1000 && *mean > 1000.0 && *dev > 1000.0) {
174 static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
177 const char *ddir_str[] = { "read ", "write" };
178 unsigned long min, max;
179 unsigned long long bw, iops;
181 char *io_p, *bw_p, *iops_p;
183 if (!ts->runtime[ddir])
186 bw = ts->io_bytes[ddir] / ts->runtime[ddir];
187 iops = (1000 * ts->total_io_u[ddir]) / ts->runtime[ddir];
188 io_p = num2str(ts->io_bytes[ddir] >> 10, 6, 1000, 1);
189 bw_p = num2str(bw, 6, 1000, 1);
190 iops_p = num2str(iops, 6, 1, 0);
192 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]);
198 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) {
199 const char *base = "(usec)";
202 if (!usec_to_msec(&min, &max, &mean, &dev))
205 minp = num2str(min, 6, 1, 0);
206 maxp = num2str(max, 6, 1, 0);
208 log_info(" slat %s: min=%s, max=%s, avg=%5.02f, stdev=%5.02f\n", base, minp, maxp, mean, dev);
213 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) {
214 const char *base = "(usec)";
217 if (!usec_to_msec(&min, &max, &mean, &dev))
220 minp = num2str(min, 6, 1, 0);
221 maxp = num2str(max, 6, 1, 0);
223 log_info(" clat %s: min=%s, max=%s, avg=%5.02f, stdev=%5.02f\n", base, minp, maxp, mean, dev);
228 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
231 p_of_agg = mean * 100 / (double) rs->agg[ddir];
232 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);
236 static void show_lat(double *io_u_lat, int nr, const char **ranges,
239 int new_line = 1, i, line = 0;
241 for (i = 0; i < nr; i++) {
242 if (io_u_lat[i] <= 0.0)
247 log_info(" lat (%s): ", msg);
253 log_info("%s%3.2f%%", ranges[i], io_u_lat[i]);
260 static void show_lat_u(double *io_u_lat_u)
262 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
263 "250=", "500=", "750=", "1000=", };
265 show_lat(io_u_lat_u, FIO_IO_U_LAT_U_NR, ranges, "usec");
268 static void show_lat_m(double *io_u_lat_m)
270 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
271 "250=", "500=", "750=", "1000=", "2000=",
274 show_lat(io_u_lat_m, FIO_IO_U_LAT_M_NR, ranges, "msec");
277 static void show_latencies(double *io_u_lat_u, double *io_u_lat_m)
279 show_lat_u(io_u_lat_u);
281 show_lat_m(io_u_lat_m);
285 static void show_thread_status(struct thread_stat *ts,
286 struct group_run_stats *rs)
288 double usr_cpu, sys_cpu;
289 unsigned long runtime;
290 double io_u_dist[FIO_IO_U_MAP_NR];
291 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
292 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
294 if (!(ts->io_bytes[0] + ts->io_bytes[1]))
298 log_info("%s: (groupid=%d, jobs=%d): err=%2d: pid=%d\n", ts->name, ts->groupid, ts->members, ts->error, ts->pid);
300 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);
303 log_info(" Description : [%s]\n", ts->description);
305 if (ts->io_bytes[DDIR_READ])
306 show_ddir_status(rs, ts, DDIR_READ);
307 if (ts->io_bytes[DDIR_WRITE])
308 show_ddir_status(rs, ts, DDIR_WRITE);
310 runtime = ts->total_run_time;
312 double runt = (double) runtime;
314 usr_cpu = (double) ts->usr_time * 100 / runt;
315 sys_cpu = (double) ts->sys_time * 100 / runt;
321 log_info(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, ts->ctx);
323 stat_calc_dist(ts, io_u_dist);
324 stat_calc_lat_u(ts, io_u_lat_u);
325 stat_calc_lat_m(ts, io_u_lat_m);
327 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]);
328 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]);
330 show_latencies(io_u_lat_u, io_u_lat_m);
333 static void show_ddir_status_terse(struct thread_stat *ts,
334 struct group_run_stats *rs, int ddir)
336 unsigned long min, max;
337 unsigned long long bw;
341 if (ts->runtime[ddir])
342 bw = ts->io_bytes[ddir] / ts->runtime[ddir];
344 log_info(";%llu;%llu;%lu", ts->io_bytes[ddir] >> 10, bw, ts->runtime[ddir]);
346 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
347 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
349 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
351 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
352 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
354 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
356 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
359 p_of_agg = mean * 100 / (double) rs->agg[ddir];
360 log_info(";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
362 log_info(";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0);
366 static void show_thread_status_terse(struct thread_stat *ts,
367 struct group_run_stats *rs)
369 double io_u_dist[FIO_IO_U_MAP_NR];
370 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
371 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
372 double usr_cpu, sys_cpu;
375 log_info("%s;%d;%d", ts->name, ts->groupid, ts->error);
377 show_ddir_status_terse(ts, rs, 0);
378 show_ddir_status_terse(ts, rs, 1);
380 if (ts->total_run_time) {
381 double runt = (double) ts->total_run_time;
383 usr_cpu = (double) ts->usr_time * 100 / runt;
384 sys_cpu = (double) ts->sys_time * 100 / runt;
390 log_info(";%f%%;%f%%;%lu", usr_cpu, sys_cpu, ts->ctx);
392 stat_calc_dist(ts, io_u_dist);
393 stat_calc_lat_u(ts, io_u_lat_u);
394 stat_calc_lat_m(ts, io_u_lat_m);
396 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]);
398 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
399 log_info(";%3.2f%%", io_u_lat_u[i]);
400 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
401 log_info(";%3.2f%%", io_u_lat_m[i]);
405 log_info(";%s", ts->description);
410 static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr)
414 dst->min_val = min(dst->min_val, src->min_val);
415 dst->max_val = max(dst->max_val, src->max_val);
416 dst->samples += src->samples;
419 * Needs a new method for calculating stddev, we cannot just
420 * average them we do below for nr > 1
426 mean = ((src->mean * (double) (nr - 1)) + dst->mean) / ((double) nr);
427 S = ((src->S * (double) (nr - 1)) + dst->S) / ((double) nr);
434 void show_run_stats(void)
436 struct group_run_stats *runstats, *rs;
437 struct thread_data *td;
438 struct thread_stat *threadstats, *ts;
439 int i, j, k, l, nr_ts, last_ts, idx;
441 runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
443 for (i = 0; i < groupid + 1; i++) {
446 memset(rs, 0, sizeof(*rs));
447 rs->min_bw[0] = rs->min_run[0] = ~0UL;
448 rs->min_bw[1] = rs->min_run[1] = ~0UL;
452 * find out how many threads stats we need. if group reporting isn't
453 * enabled, it's one-per-td.
458 if (!td->o.group_reporting) {
462 if (last_ts == td->groupid)
465 last_ts = td->groupid;
469 threadstats = malloc(nr_ts * sizeof(struct thread_stat));
471 for (i = 0; i < nr_ts; i++) {
472 ts = &threadstats[i];
474 memset(ts, 0, sizeof(*ts));
475 for (j = 0; j <= DDIR_WRITE; j++) {
476 ts->clat_stat[j].min_val = -1UL;
477 ts->slat_stat[j].min_val = -1UL;
478 ts->bw_stat[j].min_val = -1UL;
487 if (idx && (!td->o.group_reporting ||
488 (td->o.group_reporting && last_ts != td->groupid))) {
493 last_ts = td->groupid;
495 ts = &threadstats[j];
500 if (ts->groupid == -1) {
502 * These are per-group shared already
504 ts->name = td->o.name;
505 ts->description = td->o.description;
506 ts->groupid = td->groupid;
509 * first pid in group, not very useful...
514 if (td->error && !ts->error) {
515 ts->error = td->error;
516 ts->verror = td->verror;
519 for (l = 0; l <= DDIR_WRITE; l++) {
520 sum_stat(&ts->clat_stat[l], &td->ts.clat_stat[l], idx);
521 sum_stat(&ts->slat_stat[l], &td->ts.slat_stat[l], idx);
522 sum_stat(&ts->bw_stat[l], &td->ts.bw_stat[l], idx);
524 ts->stat_io_bytes[l] += td->ts.stat_io_bytes[l];
525 ts->io_bytes[l] += td->ts.io_bytes[l];
527 if (ts->runtime[l] < td->ts.runtime[l])
528 ts->runtime[l] = td->ts.runtime[l];
531 ts->usr_time += td->ts.usr_time;
532 ts->sys_time += td->ts.sys_time;
533 ts->ctx += td->ts.ctx;
535 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
536 ts->io_u_map[k] += td->ts.io_u_map[k];
537 for (k = 0; k < FIO_IO_U_LAT_U_NR; k++)
538 ts->io_u_lat_u[k] += td->ts.io_u_lat_u[k];
539 for (k = 0; k < FIO_IO_U_LAT_M_NR; k++)
540 ts->io_u_lat_m[k] += td->ts.io_u_lat_m[k];
543 for (k = 0; k <= DDIR_WRITE; k++) {
544 ts->total_io_u[k] += td->ts.total_io_u[k];
545 ts->short_io_u[k] += td->ts.short_io_u[k];
548 ts->total_run_time += td->ts.total_run_time;
551 for (i = 0; i < nr_ts; i++) {
552 unsigned long long bw;
554 ts = &threadstats[i];
555 rs = &runstats[ts->groupid];
557 for (j = 0; j <= DDIR_WRITE; j++) {
560 if (ts->runtime[j] < rs->min_run[j] || !rs->min_run[j])
561 rs->min_run[j] = ts->runtime[j];
562 if (ts->runtime[j] > rs->max_run[j])
563 rs->max_run[j] = ts->runtime[j];
567 bw = ts->io_bytes[j] / (unsigned long long) ts->runtime[j];
568 if (bw < rs->min_bw[j])
570 if (bw > rs->max_bw[j])
573 rs->io_kb[j] += ts->io_bytes[j] >> 10;
577 for (i = 0; i < groupid + 1; i++) {
581 rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0];
583 rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1];
587 * don't overwrite last signal output
592 for (i = 0; i < nr_ts; i++) {
593 ts = &threadstats[i];
594 rs = &runstats[ts->groupid];
597 show_thread_status_terse(ts, rs);
599 show_thread_status(ts, rs);
603 for (i = 0; i < groupid + 1; i++)
604 show_group_stats(&runstats[i], i);
613 static inline void add_stat_sample(struct io_stat *is, unsigned long data)
618 if (data > is->max_val)
620 if (data < is->min_val)
623 delta = val - is->mean;
625 is->mean += delta / (is->samples + 1.0);
626 is->S += delta * (val - is->mean);
632 static void __add_log_sample(struct io_log *iolog, unsigned long val,
633 enum fio_ddir ddir, unsigned long time)
635 if (iolog->nr_samples == iolog->max_samples) {
636 int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
638 iolog->log = realloc(iolog->log, new_size);
639 iolog->max_samples <<= 1;
642 iolog->log[iolog->nr_samples].val = val;
643 iolog->log[iolog->nr_samples].time = time;
644 iolog->log[iolog->nr_samples].ddir = ddir;
648 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
649 unsigned long val, enum fio_ddir ddir)
651 __add_log_sample(iolog, val, ddir, mtime_since_now(&td->epoch));
654 void add_agg_sample(unsigned long val, enum fio_ddir ddir)
656 struct io_log *iolog = agg_io_log[ddir];
658 __add_log_sample(iolog, val, ddir, mtime_since_genesis());
661 void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
664 struct thread_stat *ts = &td->ts;
666 add_stat_sample(&ts->clat_stat[ddir], usec);
669 add_log_sample(td, ts->clat_log, usec, ddir);
672 void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
675 struct thread_stat *ts = &td->ts;
677 add_stat_sample(&ts->slat_stat[ddir], usec);
680 add_log_sample(td, ts->slat_log, usec, ddir);
683 void add_bw_sample(struct thread_data *td, enum fio_ddir ddir,
686 struct thread_stat *ts = &td->ts;
687 unsigned long spent = mtime_since(&ts->stat_sample_time[ddir], t);
690 if (spent < td->o.bw_avg_time)
693 rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) / spent;
694 add_stat_sample(&ts->bw_stat[ddir], rate);
697 add_log_sample(td, ts->bw_log, rate, ddir);
699 fio_gettime(&ts->stat_sample_time[ddir], NULL);
700 ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir];