X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=stat.c;h=3a1546778eeca8dbf21268f276c16616c8b6f7d2;hp=731b3036544de363e9db4a82084cce1316203d1e;hb=197574e43bea5ac00f7b64d4731173c177041de0;hpb=690089990d051d86b4ef2b6fd5c1972c0dd4897b diff --git a/stat.c b/stat.c index 731b3036..3a154677 100644 --- a/stat.c +++ b/stat.c @@ -11,6 +11,47 @@ static struct itimerval itimer; static struct list_head disk_list = LIST_HEAD_INIT(disk_list); +static dev_t last_dev; + +/* + * Cheesy number->string conversion, complete with carry rounding error. + */ +static char *num2str(unsigned long num, int maxlen, int base) +{ + /* + * could be passed in for 10^3 base, but every caller expects + * 2^10 base right now. + */ + const unsigned int thousand = 1024; + char postfix[] = { 'K', 'M', 'G', 'P', 'E' }; + char *buf; + int i; + + buf = malloc(128); + + for (i = 0; base > 1; i++) + base /= thousand; + + do { + int len, carry = 0; + + len = sprintf(buf, "%'lu", num); + if (len <= maxlen) { + buf[len] = postfix[i]; + buf[len + 1] = '\0'; + return buf; + } + + if ((num % thousand) >= (thousand / 2)) + carry = 1; + + num /= thousand; + num += carry; + i++; + } while (i <= 5); + + return buf; +} static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus) { @@ -94,14 +135,26 @@ static int disk_util_exists(dev_t dev) static void disk_util_add(dev_t dev, char *path) { - struct disk_util *du = malloc(sizeof(*du)); + struct disk_util *du, *__du; + struct list_head *entry; + du = malloc(sizeof(*du)); memset(du, 0, sizeof(*du)); INIT_LIST_HEAD(&du->list); sprintf(du->path, "%s/stat", path); du->name = strdup(basename(path)); du->dev = dev; + list_for_each(entry, &disk_list) { + __du = list_entry(entry, struct disk_util, list); + + if (!strcmp(du->name, __du->name)) { + free(du->name); + free(du); + return; + } + } + fio_gettime(&du->time, NULL); get_io_ticks(du, &du->last_dus); @@ -193,7 +246,7 @@ void init_disk_util(struct thread_data *td) dev_t dev; char *p; - if (!td->do_disk_util) + if (!td->do_disk_util || (td->io_ops->flags & FIO_DISKLESSIO)) return; /* @@ -221,6 +274,17 @@ void init_disk_util(struct thread_data *td) if (disk_util_exists(dev)) return; + + /* + * for an fs without a device, we will repeatedly stat through + * sysfs which can take oodles of time for thousands of files. so + * cache the last lookup and compare with that before going through + * everything again. + */ + if (dev == last_dev) + return; + + last_dev = dev; sprintf(foo, "/sys/block"); if (!find_block_dir(dev, foo)) @@ -258,20 +322,21 @@ void disk_util_timer_arm(void) void update_rusage_stat(struct thread_data *td) { - getrusage(RUSAGE_SELF, &td->ru_end); + struct thread_stat *ts = &td->ts; - td->usr_time += mtime_since(&td->ru_start.ru_utime, &td->ru_end.ru_utime); - td->sys_time += mtime_since(&td->ru_start.ru_stime, &td->ru_end.ru_stime); - td->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw); + getrusage(RUSAGE_SELF, &ts->ru_end); + ts->usr_time += mtime_since(&ts->ru_start.ru_utime, &ts->ru_end.ru_utime); + ts->sys_time += mtime_since(&ts->ru_start.ru_stime, &ts->ru_end.ru_stime); + ts->ctx += ts->ru_end.ru_nvcsw + ts->ru_end.ru_nivcsw - (ts->ru_start.ru_nvcsw + ts->ru_start.ru_nivcsw); - memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end)); + memcpy(&ts->ru_start, &ts->ru_end, sizeof(ts->ru_end)); } static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max, double *mean, double *dev) { - double n; + double n = is->samples; if (is->samples == 0) return 0; @@ -280,20 +345,40 @@ static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max, *max = is->max_val; n = (double) is->samples; - *mean = (double) is->val / n; - *dev = sqrt(((double) is->val_sq - (*mean * *mean) / n) / (n - 1)); + *mean = is->mean; + + if (n > 1.0) + *dev = sqrt(is->S / (n - 1.0)); + else + *dev = -1.0; return 1; } static void show_group_stats(struct group_run_stats *rs, int id) { + char *p1, *p2, *p3, *p4; + const char *ddir_str[] = { " READ", " WRITE" }; + int i; + fprintf(f_out, "\nRun status group %d (all jobs):\n", id); - if (rs->max_run[DDIR_READ]) - fprintf(f_out, " READ: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[0] >> 10, rs->agg[0], rs->min_bw[0], rs->max_bw[0], rs->min_run[0], rs->max_run[0]); - if (rs->max_run[DDIR_WRITE]) - fprintf(f_out, " WRITE: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[1] >> 10, rs->agg[1], rs->min_bw[1], rs->max_bw[1], rs->min_run[1], rs->max_run[1]); + for (i = 0; i <= DDIR_WRITE; i++) { + if (!rs->max_run[i]) + continue; + + p1 = num2str(rs->io_kb[i], 6, 1); + p2 = num2str(rs->agg[i], 6, 1); + p3 = num2str(rs->min_bw[i], 6, 1); + p4 = num2str(rs->max_bw[i], 6, 1); + + fprintf(f_out, "%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]); + + free(p1); + free(p2); + free(p3); + free(p4); + } } static void show_disk_util(void) @@ -327,64 +412,114 @@ static void show_disk_util(void) } } -static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs, +static void stat_calc_dist(struct thread_stat *ts, double *io_u_dist) +{ + int i; + + /* + * Do depth distribution calculations + */ + for (i = 0; i < FIO_IO_U_MAP_NR; i++) { + io_u_dist[i] = (double) ts->io_u_map[i] / (double) ts->total_io_u; + io_u_dist[i] *= 100.0; + } +} + +static void stat_calc_lat(struct thread_stat *ts, double *io_u_lat) +{ + int i; + + /* + * Do latency distribution calculations + */ + for (i = 0; i < FIO_IO_U_LAT_NR; i++) { + io_u_lat[i] = (double) ts->io_u_lat[i] / (double) ts->total_io_u; + io_u_lat[i] *= 100.0; + } +} + +static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts, int ddir) { const char *ddir_str[] = { "read ", "write" }; unsigned long min, max; unsigned long long bw; double mean, dev; + char *io_p, *bw_p; - if (!td->runtime[ddir]) + if (!ts->runtime[ddir]) return; - bw = td->io_bytes[ddir] / td->runtime[ddir]; - fprintf(f_out, " %s: io=%6lluMiB, bw=%6lluKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]); + bw = ts->io_bytes[ddir] / ts->runtime[ddir]; + io_p = num2str(ts->io_bytes[ddir] >> 10, 6, 1); + bw_p = num2str(bw, 6, 1); + + fprintf(f_out, " %s: io=%siB, bw=%siB/s, runt=%6lumsec\n", ddir_str[ddir], io_p, bw_p, ts->runtime[ddir]); - if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev)) - fprintf(f_out, " slat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev); + free(io_p); + free(bw_p); - if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev)) - fprintf(f_out, " clat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev); + if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) + fprintf(f_out, " slat (msec): min=%5lu, max=%5lu, avg=%5.02f, stdev=%5.02f\n", min, max, mean, dev); - if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) { + if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) + fprintf(f_out, " clat (msec): min=%5lu, max=%5lu, avg=%5.02f, stdev=%5.02f\n", min, max, mean, dev); + + if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) { double p_of_agg; p_of_agg = mean * 100 / (double) rs->agg[ddir]; - fprintf(f_out, " bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, dev=%5.02f\n", min, max, p_of_agg, mean, dev); + fprintf(f_out, " bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg, mean, dev); } } -static void show_thread_status(struct thread_data *td, +static void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs) { double usr_cpu, sys_cpu; unsigned long runtime; + double io_u_dist[FIO_IO_U_MAP_NR]; + double io_u_lat[FIO_IO_U_LAT_NR]; - if (!(td->io_bytes[0] + td->io_bytes[1]) && !td->error) + if (!(ts->io_bytes[0] + ts->io_bytes[1])) return; - fprintf(f_out, "%s: (groupid=%d): err=%2d:\n",td->name, td->groupid, td->error); + if (!ts->error) + fprintf(f_out, "%s: (groupid=%d, jobs=%d): err=%2d: pid=%d\n", ts->name, ts->groupid, ts->members, ts->error, ts->pid); + else + fprintf(f_out, "%s: (groupid=%d, jobs=%d): err=%2d (%s): pid=%d\n", ts->name, ts->groupid, ts->members, ts->error, ts->verror, ts->pid); - show_ddir_status(td, rs, td->ddir); - if (td->io_bytes[td->ddir ^ 1]) - show_ddir_status(td, rs, td->ddir ^ 1); + if (ts->description) + fprintf(f_out, " Description : [%s]\n", ts->description); - runtime = mtime_since(&td->epoch, &td->end_time); + if (ts->io_bytes[DDIR_READ]) + show_ddir_status(rs, ts, DDIR_READ); + if (ts->io_bytes[DDIR_WRITE]) + show_ddir_status(rs, ts, DDIR_WRITE); + + runtime = ts->total_run_time; if (runtime) { - double runt = runtime; + double runt = (double) runtime; - usr_cpu = (double) td->usr_time * 100 / runt; - sys_cpu = (double) td->sys_time * 100 / runt; + usr_cpu = (double) ts->usr_time * 100 / runt; + sys_cpu = (double) ts->sys_time * 100 / runt; } else { usr_cpu = 0; sys_cpu = 0; } - fprintf(f_out, " cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx); + fprintf(f_out, " cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, ts->ctx); + + stat_calc_dist(ts, io_u_dist); + stat_calc_lat(ts, io_u_lat); + + fprintf(f_out, " 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]); + + fprintf(f_out, " lat (msec): 2=%3.1f%%, 4=%3.1f%%, 10=%3.1f%%, 20=%3.1f%%, 50=%3.1f%%, 100=%3.1f%%\n", io_u_lat[0], io_u_lat[1], io_u_lat[2], io_u_lat[3], io_u_lat[4], io_u_lat[5]); + fprintf(f_out, " lat (msec): 250=%3.1f%%, 500=%3.1f%%, 750=%3.1f%%, 1000=%3.1f%%, >=2000=%3.1f%%\n", io_u_lat[6], io_u_lat[7], io_u_lat[8], io_u_lat[9], io_u_lat[10]); } -static void show_ddir_status_terse(struct thread_data *td, +static void show_ddir_status_terse(struct thread_stat *ts, struct group_run_stats *rs, int ddir) { unsigned long min, max; @@ -392,60 +527,99 @@ static void show_ddir_status_terse(struct thread_data *td, double mean, dev; bw = 0; - if (td->runtime[ddir]) - bw = td->io_bytes[ddir] / td->runtime[ddir]; + if (ts->runtime[ddir]) + bw = ts->io_bytes[ddir] / ts->runtime[ddir]; - fprintf(f_out, ",%llu,%llu,%lu", td->io_bytes[ddir] >> 10, bw, td->runtime[ddir]); + fprintf(f_out, ";%llu;%llu;%lu", ts->io_bytes[ddir] >> 10, bw, ts->runtime[ddir]); - if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev)) - fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev); + if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) + fprintf(f_out, ";%lu;%lu;%f;%f", min, max, mean, dev); else - fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0); + fprintf(f_out, ";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0); - if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev)) - fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev); + if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) + fprintf(f_out, ";%lu;%lu;%f;%f", min, max, mean, dev); else - fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0); + fprintf(f_out, ";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0); - if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) { + if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) { double p_of_agg; p_of_agg = mean * 100 / (double) rs->agg[ddir]; - fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", min, max, p_of_agg, mean, dev); + fprintf(f_out, ";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev); } else - fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", 0UL, 0UL, 0.0, 0.0, 0.0); - + fprintf(f_out, ";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0); } -static void show_thread_status_terse(struct thread_data *td, +static void show_thread_status_terse(struct thread_stat *ts, struct group_run_stats *rs) { + double io_u_dist[FIO_IO_U_MAP_NR]; + double io_u_lat[FIO_IO_U_LAT_NR]; double usr_cpu, sys_cpu; - fprintf(f_out, "%s,%d,%d",td->name, td->groupid, td->error); + fprintf(f_out, "%s;%d;%d", ts->name, ts->groupid, ts->error); - show_ddir_status_terse(td, rs, 0); - show_ddir_status_terse(td, rs, 1); + show_ddir_status_terse(ts, rs, 0); + show_ddir_status_terse(ts, rs, 1); - if (td->runtime[0] + td->runtime[1]) { - double runt = td->runtime[0] + td->runtime[1]; + if (ts->total_run_time) { + double runt = (double) ts->total_run_time; - usr_cpu = (double) td->usr_time * 100 / runt; - sys_cpu = (double) td->sys_time * 100 / runt; + usr_cpu = (double) ts->usr_time * 100 / runt; + sys_cpu = (double) ts->sys_time * 100 / runt; } else { usr_cpu = 0; sys_cpu = 0; } - fprintf(f_out, ",%f%%,%f%%,%lu\n", usr_cpu, sys_cpu, td->ctx); + fprintf(f_out, ";%f%%;%f%%;%lu", usr_cpu, sys_cpu, ts->ctx); + + stat_calc_dist(ts, io_u_dist); + stat_calc_lat(ts, io_u_lat); + + fprintf(f_out, ";%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]); + + fprintf(f_out, ";%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%\n", io_u_lat[0], io_u_lat[1], io_u_lat[2], io_u_lat[3], io_u_lat[4], io_u_lat[5]); + fprintf(f_out, ";%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%", io_u_lat[6], io_u_lat[7], io_u_lat[8], io_u_lat[9], io_u_lat[10]); + + if (ts->description) + fprintf(f_out, ";%s", ts->description); + + fprintf(f_out, "\n"); +} + +static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr) +{ + double mean, S; + + dst->min_val = min(dst->min_val, src->min_val); + dst->max_val = max(dst->max_val, src->max_val); + dst->samples += src->samples; + + /* + * Needs a new method for calculating stddev, we cannot just + * average them we do below for nr > 1 + */ + if (nr == 1) { + mean = src->mean; + S = src->S; + } else { + mean = ((src->mean * (double) (nr - 1)) + dst->mean) / ((double) nr); + S = ((src->S * (double) (nr - 1)) + dst->S) / ((double) nr); + } + + dst->mean = mean; + dst->S = S; } void show_run_stats(void) { struct group_run_stats *runstats, *rs; struct thread_data *td; - int i; + struct thread_stat *threadstats, *ts; + int i, j, k, l, nr_ts, last_ts, idx; runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1)); @@ -457,42 +631,130 @@ void show_run_stats(void) rs->min_bw[1] = rs->min_run[1] = ~0UL; } + /* + * find out how many threads stats we need. if group reporting isn't + * enabled, it's one-per-td. + */ + nr_ts = 0; + last_ts = -1; for_each_td(td, i) { - unsigned long long rbw, wbw; + if (!td->group_reporting) { + nr_ts++; + continue; + } + if (last_ts == td->groupid) + continue; + + last_ts = td->groupid; + nr_ts++; + } + + threadstats = malloc(nr_ts * sizeof(struct thread_stat)); + + for (i = 0; i < nr_ts; i++) { + ts = &threadstats[i]; + + memset(ts, 0, sizeof(*ts)); + for (j = 0; j < 2; j++) { + ts->clat_stat[j].min_val = -1UL; + ts->slat_stat[j].min_val = -1UL; + ts->bw_stat[j].min_val = -1UL; + } + } + + j = 0; + last_ts = -1; + idx = 0; + for_each_td(td, i) { + ts = &threadstats[j]; + + idx++; + ts->members++; + + if (!ts->groupid) { + /* + * These are per-group shared already + */ + ts->name = td->name; + ts->description = td->description; + ts->groupid = td->groupid; + + /* + * first pid in group, not very useful... + */ + ts->pid = td->pid; + } + + if (td->error && !ts->error) { + ts->error = td->error; + ts->verror = td->verror; + } - if (td->error) { - fprintf(f_out, "%s: %s\n", td->name, td->verror); + for (l = 0; l < 2; l++) { + sum_stat(&ts->clat_stat[l], &td->ts.clat_stat[l], idx); + sum_stat(&ts->slat_stat[l], &td->ts.slat_stat[l], idx); + sum_stat(&ts->bw_stat[l], &td->ts.bw_stat[l], idx); + + ts->stat_io_bytes[l] += td->ts.stat_io_bytes[l]; + ts->io_bytes[l] += td->ts.io_bytes[l]; + + if (ts->runtime[l] < td->ts.runtime[l]) + ts->runtime[l] = td->ts.runtime[l]; + } + + ts->usr_time += td->ts.usr_time; + ts->sys_time += td->ts.sys_time; + ts->ctx += td->ts.ctx; + + for (k = 0; k < FIO_IO_U_MAP_NR; k++) + ts->io_u_map[k] += td->ts.io_u_map[k]; + for (k = 0; k < FIO_IO_U_LAT_NR; k++) + ts->io_u_lat[k] += td->ts.io_u_lat[k]; + + ts->total_io_u += td->ts.total_io_u; + + ts->total_run_time += td->ts.total_run_time; + + if (!td->group_reporting) { + idx = 0; + j++; continue; } + if (last_ts == td->groupid) + continue; + + if (last_ts != -1) { + idx = 0; + j++; + } - rs = &runstats[td->groupid]; - - if (td->runtime[0] < rs->min_run[0] || !rs->min_run[0]) - rs->min_run[0] = td->runtime[0]; - if (td->runtime[0] > rs->max_run[0]) - rs->max_run[0] = td->runtime[0]; - if (td->runtime[1] < rs->min_run[1] || !rs->min_run[1]) - rs->min_run[1] = td->runtime[1]; - if (td->runtime[1] > rs->max_run[1]) - rs->max_run[1] = td->runtime[1]; - - rbw = wbw = 0; - if (td->runtime[0]) - rbw = td->io_bytes[0] / (unsigned long long) td->runtime[0]; - if (td->runtime[1]) - wbw = td->io_bytes[1] / (unsigned long long) td->runtime[1]; - - if (rbw < rs->min_bw[0]) - rs->min_bw[0] = rbw; - if (wbw < rs->min_bw[1]) - rs->min_bw[1] = wbw; - if (rbw > rs->max_bw[0]) - rs->max_bw[0] = rbw; - if (wbw > rs->max_bw[1]) - rs->max_bw[1] = wbw; - - rs->io_kb[0] += td->io_bytes[0] >> 10; - rs->io_kb[1] += td->io_bytes[1] >> 10; + last_ts = td->groupid; + } + + for (i = 0; i < nr_ts; i++) { + unsigned long long bw; + + ts = &threadstats[i]; + rs = &runstats[ts->groupid]; + + for (j = 0; j < 2; j++) { + if (!ts->runtime[j]) + continue; + if (ts->runtime[j] < rs->min_run[j] || !rs->min_run[j]) + rs->min_run[j] = ts->runtime[j]; + if (ts->runtime[j] > rs->max_run[j]) + rs->max_run[j] = ts->runtime[j]; + + bw = 0; + if (ts->runtime[j]) + bw = ts->io_bytes[j] / (unsigned long long) ts->runtime[j]; + if (bw < rs->min_bw[j]) + rs->min_bw[j] = bw; + if (bw > rs->max_bw[j]) + rs->max_bw[j] = bw; + + rs->io_kb[j] += ts->io_bytes[j] >> 10; + } } for (i = 0; i < groupid + 1; i++) { @@ -510,13 +772,14 @@ void show_run_stats(void) if (!terse_output) printf("\n"); - for_each_td(td, i) { - rs = &runstats[td->groupid]; + for (i = 0; i < nr_ts; i++) { + ts = &threadstats[i]; + rs = &runstats[ts->groupid]; if (terse_output) - show_thread_status_terse(td, rs); + show_thread_status_terse(ts, rs); else - show_thread_status(td, rs); + show_thread_status(ts, rs); } if (!terse_output) { @@ -527,22 +790,28 @@ void show_run_stats(void) } free(runstats); + free(threadstats); } -static inline void add_stat_sample(struct io_stat *is, unsigned long val) +static inline void add_stat_sample(struct io_stat *is, unsigned long data) { - if (val > is->max_val) - is->max_val = val; - if (val < is->min_val) - is->min_val = val; + double val = data; + double delta; + + if (data > is->max_val) + is->max_val = data; + if (data < is->min_val) + is->min_val = data; + + delta = val - is->mean; + is->mean += delta / (is->samples + 1.0); + is->S += delta * (val - is->mean); - is->val += val; - is->val_sq += val * val; is->samples++; } -static void add_log_sample(struct thread_data *td, struct io_log *iolog, - unsigned long val, int ddir) +static void __add_log_sample(struct io_log *iolog, unsigned long val, + enum fio_ddir ddir, unsigned long time) { if (iolog->nr_samples == iolog->max_samples) { int new_size = sizeof(struct io_sample) * iolog->max_samples*2; @@ -552,43 +821,62 @@ static void add_log_sample(struct thread_data *td, struct io_log *iolog, } iolog->log[iolog->nr_samples].val = val; - iolog->log[iolog->nr_samples].time = mtime_since_now(&td->epoch); + iolog->log[iolog->nr_samples].time = time; iolog->log[iolog->nr_samples].ddir = ddir; iolog->nr_samples++; } -void add_clat_sample(struct thread_data *td, int ddir, unsigned long msec) +static void add_log_sample(struct thread_data *td, struct io_log *iolog, + unsigned long val, enum fio_ddir ddir) +{ + __add_log_sample(iolog, val, ddir, mtime_since_now(&td->epoch)); +} + +void add_agg_sample(unsigned long val, enum fio_ddir ddir) { - add_stat_sample(&td->clat_stat[ddir], msec); + struct io_log *iolog = agg_io_log[ddir]; - if (td->clat_log) - add_log_sample(td, td->clat_log, msec, ddir); + __add_log_sample(iolog, val, ddir, mtime_since_genesis()); } -void add_slat_sample(struct thread_data *td, int ddir, unsigned long msec) +void add_clat_sample(struct thread_data *td, enum fio_ddir ddir, + unsigned long msec) { - add_stat_sample(&td->slat_stat[ddir], msec); + struct thread_stat *ts = &td->ts; + + add_stat_sample(&ts->clat_stat[ddir], msec); - if (td->slat_log) - add_log_sample(td, td->slat_log, msec, ddir); + if (ts->clat_log) + add_log_sample(td, ts->clat_log, msec, ddir); } -void add_bw_sample(struct thread_data *td, int ddir, struct timeval *t) +void add_slat_sample(struct thread_data *td, enum fio_ddir ddir, + unsigned long msec) { - unsigned long spent = mtime_since(&td->stat_sample_time[ddir], t); + struct thread_stat *ts = &td->ts; + + add_stat_sample(&ts->slat_stat[ddir], msec); + + if (ts->slat_log) + add_log_sample(td, ts->slat_log, msec, ddir); +} + +void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, + struct timeval *t) +{ + struct thread_stat *ts = &td->ts; + unsigned long spent = mtime_since(&ts->stat_sample_time[ddir], t); unsigned long rate; if (spent < td->bw_avg_time) return; - rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) / spent; - add_stat_sample(&td->bw_stat[ddir], rate); + rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) / spent; + add_stat_sample(&ts->bw_stat[ddir], rate); - if (td->bw_log) - add_log_sample(td, td->bw_log, rate, ddir); + if (ts->bw_log) + add_log_sample(td, ts->bw_log, rate, ddir); - fio_gettime(&td->stat_sample_time[ddir], NULL); - td->stat_io_bytes[ddir] = td->this_io_bytes[ddir]; + fio_gettime(&ts->stat_sample_time[ddir], NULL); + ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir]; } - -