X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=stat.c;h=6c652d6e4f8708039c7074d833471d3e0ddd0c0b;hb=4e5b8fb8e2d989789e18b31f9aed969c8fa43cdb;hp=a596388a41c5b00f820fb2f79f69436dfecd9aa2;hpb=cda99fa0fcca41fc0d54ecaa2fd600162a8de5d3;p=fio.git diff --git a/stat.c b/stat.c index a596388a..6c652d6e 100644 --- a/stat.c +++ b/stat.c @@ -9,23 +9,158 @@ #include "fio.h" #include "diskutil.h" +#include "ieee754.h" void update_rusage_stat(struct thread_data *td) { struct thread_stat *ts = &td->ts; - getrusage(RUSAGE_SELF, &ts->ru_end); + getrusage(RUSAGE_SELF, &td->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); - ts->minf += ts->ru_end.ru_minflt - ts->ru_start.ru_minflt; - ts->majf += ts->ru_end.ru_majflt - ts->ru_start.ru_majflt; + ts->usr_time += mtime_since(&td->ru_start.ru_utime, + &td->ru_end.ru_utime); + ts->sys_time += mtime_since(&td->ru_start.ru_stime, + &td->ru_end.ru_stime); + ts->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw + - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw); + ts->minf += td->ru_end.ru_minflt - td->ru_start.ru_minflt; + ts->majf += td->ru_end.ru_majflt - td->ru_start.ru_majflt; - memcpy(&ts->ru_start, &ts->ru_end, sizeof(ts->ru_end)); + memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end)); +} + +/* + * Given a latency, return the index of the corresponding bucket in + * the structure tracking percentiles. + * + * (1) find the group (and error bits) that the value (latency) + * belongs to by looking at its MSB. (2) find the bucket number in the + * group by looking at the index bits. + * + */ +static unsigned int plat_val_to_idx(unsigned int val) +{ + unsigned int msb, error_bits, base, offset, idx; + + /* Find MSB starting from bit 0 */ + if (val == 0) + msb = 0; + else + msb = (sizeof(val)*8) - __builtin_clz(val) - 1; + + /* + * MSB <= (FIO_IO_U_PLAT_BITS-1), cannot be rounded off. Use + * all bits of the sample as index + */ + if (msb <= FIO_IO_U_PLAT_BITS) + return val; + + /* Compute the number of error bits to discard*/ + error_bits = msb - FIO_IO_U_PLAT_BITS; + + /* Compute the number of buckets before the group */ + base = (error_bits + 1) << FIO_IO_U_PLAT_BITS; + + /* + * Discard the error bits and apply the mask to find the + * index for the buckets in the group + */ + offset = (FIO_IO_U_PLAT_VAL - 1) & (val >> error_bits); + + /* Make sure the index does not exceed (array size - 1) */ + idx = (base + offset) < (FIO_IO_U_PLAT_NR - 1)? + (base + offset) : (FIO_IO_U_PLAT_NR - 1); + + return idx; +} + +/* + * Convert the given index of the bucket array to the value + * represented by the bucket + */ +static unsigned int plat_idx_to_val(unsigned int idx) +{ + unsigned int error_bits, k, base; + + assert(idx < FIO_IO_U_PLAT_NR); + + /* MSB <= (FIO_IO_U_PLAT_BITS-1), cannot be rounded off. Use + * all bits of the sample as index */ + if (idx < (FIO_IO_U_PLAT_VAL << 1) ) + return idx; + + /* Find the group and compute the minimum value of that group */ + error_bits = (idx >> FIO_IO_U_PLAT_BITS) -1; + base = 1 << (error_bits + FIO_IO_U_PLAT_BITS); + + /* Find its bucket number of the group */ + k = idx % FIO_IO_U_PLAT_VAL; + + /* Return the mean of the range of the bucket */ + return base + ((k + 0.5) * (1 << error_bits)); +} + +static int double_cmp(const void *a, const void *b) +{ + const fio_fp64_t fa = *(const fio_fp64_t *) a; + const fio_fp64_t fb = *(const fio_fp64_t *) b; + int cmp = 0; + + if (fa.u.f > fb.u.f) + cmp = 1; + else if (fa.u.f < fb.u.f) + cmp = -1; + + return cmp; +} + +/* + * Find and display the p-th percentile of clat + */ +static void show_clat_percentiles(unsigned int *io_u_plat, unsigned long nr, + fio_fp64_t *plist) +{ + unsigned long sum = 0; + unsigned int len, i, j = 0; + int is_last = 0; + + len = 0; + while (len < FIO_IO_U_LIST_MAX_LEN && plist[len].u.f != 0.0) + len++; + + /* + * Sort the percentile list. Note that it may already be sorted if + * we are using the default values, but since it's a short list this + * isn't a worry. Also note that this does not work for NaN values. + */ + if (len > 1) + qsort((void*)plist, len, sizeof(plist[0]), double_cmp); + + log_info(" clat percentiles (usec) :"); + + for (i = 0; i < FIO_IO_U_PLAT_NR && !is_last; i++) { + sum += io_u_plat[i]; + while (sum >= (plist[j].u.f / 100.0 * nr)) { + assert(plist[j].u.f <= 100.0); + + /* for formatting */ + if (j != 0 && (j % 4) == 0) + log_info(" "); + + /* end of the list */ + is_last = (j == len - 1); + + log_info(" %2.2fth=%u%c", plist[j], plat_idx_to_val(i), + (is_last? '\n' : ',')); + + if (is_last) + break; + + if (j % 4 == 3) /* for formatting */ + log_info("\n"); + j++; + } + } } static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max, @@ -40,23 +175,23 @@ static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max, *max = is->max_val; n = (double) is->samples; - *mean = is->mean; + *mean = is->mean.u.f; if (n > 1.0) - *dev = sqrt(is->S / (n - 1.0)); + *dev = sqrt(is->S.u.f / (n - 1.0)); else *dev = 0; return 1; } -static void show_group_stats(struct group_run_stats *rs, int id) +void show_group_stats(struct group_run_stats *rs) { char *p1, *p2, *p3, *p4; const char *ddir_str[] = { " READ", " WRITE" }; int i; - log_info("\nRun status group %d (all jobs):\n", id); + log_info("\nRun status group %d (all jobs):\n", rs->groupid); for (i = 0; i <= DDIR_WRITE; i++) { const int i2p = is_power_of_2(rs->kb_base); @@ -169,7 +304,7 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts, io_p = num2str(ts->io_bytes[ddir], 6, 1, i2p); bw_p = num2str(bw, 6, 1, i2p); - iops = (1000 * ts->total_io_u[ddir]) / runt; + iops = (1000 * (uint64_t)ts->total_io_u[ddir]) / runt; iops_p = num2str(iops, 6, 1, 0); log_info(" %s: io=%sB, bw=%sB/s, iops=%s, runt=%6llumsec\n", @@ -228,11 +363,16 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts, free(minp); free(maxp); } + if (ts->clat_percentiles) { + show_clat_percentiles(ts->io_u_plat[ddir], + ts->clat_stat[ddir].samples, + ts->percentile_list); + } if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) { double p_of_agg; p_of_agg = mean * 100 / (double) rs->agg[ddir]; - log_info(" bw (KB/s) : min=%5lu, max=%5lu, per=%3.2f%%," + log_info(" bw (KB/s) : min=%5lu, max=%5lu, per=%3.2f%%," " avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg, mean, dev); } @@ -287,8 +427,7 @@ static void show_latencies(double *io_u_lat_u, double *io_u_lat_m) log_info("\n"); } -static void show_thread_status(struct thread_stat *ts, - struct group_run_stats *rs) +void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs) { double usr_cpu, sys_cpu; unsigned long runtime; @@ -472,25 +611,35 @@ static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr) { double mean, S; + if (src->samples == 0) + return; + 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 + * Compute new mean and S after the merge + * */ if (nr == 1) { - mean = src->mean; - S = src->S; + mean = src->mean.u.f; + S = src->S.u.f; } else { - mean = ((src->mean * (double) (nr - 1)) - + dst->mean) / ((double) nr); - S = ((src->S * (double) (nr - 1)) + dst->S) / ((double) nr); + double delta = src->mean.u.f - dst->mean.u.f; + + mean = ((src->mean.u.f * src->samples) + + (dst->mean.u.f * dst->samples)) / + (dst->samples + src->samples); + + S = src->S.u.f + dst->S.u.f + pow(delta, 2.0) * + (dst->samples * src->samples) / + (dst->samples + src->samples); } - dst->mean = mean; - dst->S = S; + dst->samples += src->samples; + dst->mean.u.f = mean; + dst->S.u.f = S; } void show_run_stats(void) @@ -558,6 +707,12 @@ void show_run_stats(void) ts = &threadstats[j]; + ts->clat_percentiles = td->o.clat_percentiles; + if (td->o.overwrite_plist) + memcpy(ts->percentile_list, td->o.percentile_list, sizeof(td->o.percentile_list)); + else + memcpy(ts->percentile_list, def_percentile_list, sizeof(def_percentile_list)); + idx++; ts->members++; @@ -565,8 +720,13 @@ void show_run_stats(void) /* * These are per-group shared already */ - ts->name = td->o.name; - ts->description = td->o.description; + strncpy(ts->name, td->o.name, FIO_JOBNAME_SIZE); + if (td->o.description) + strncpy(ts->description, td->o.description, + FIO_JOBNAME_SIZE); + else + memset(ts->description, 0, FIO_JOBNAME_SIZE); + ts->groupid = td->groupid; /* @@ -588,10 +748,10 @@ void show_run_stats(void) if (!td->error && td->o.continue_on_error && td->first_error) { ts->error = td->first_error; - ts->verror = td->verror; + strcpy(ts->verror, td->verror); } else if (td->error) { ts->error = td->error; - ts->verror = td->verror; + strcpy(ts->verror, td->verror); } } @@ -601,7 +761,6 @@ void show_run_stats(void) sum_stat(&ts->lat_stat[l], &td->ts.lat_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]) @@ -631,6 +790,12 @@ void show_run_stats(void) ts->short_io_u[k] += td->ts.short_io_u[k]; } + for (k = 0; k <= DDIR_WRITE; k++) { + int m; + for (m = 0; m < FIO_IO_U_PLAT_NR; m++) + ts->io_u_plat[k][m] += td->ts.io_u_plat[k][m]; + } + ts->total_run_time += td->ts.total_run_time; ts->total_submit += td->ts.total_submit; ts->total_complete += td->ts.total_complete; @@ -690,15 +855,24 @@ void show_run_stats(void) ts = &threadstats[i]; rs = &runstats[ts->groupid]; - if (terse_output) + if (is_backend) + fio_server_send_ts(ts, rs); + else if (terse_output) show_thread_status_terse(ts, rs); else show_thread_status(ts, rs); } if (!terse_output) { - for (i = 0; i < groupid + 1; i++) - show_group_stats(&runstats[i], i); + for (i = 0; i < groupid + 1; i++) { + rs = &runstats[i]; + + rs->groupid = i; + if (is_backend) + fio_server_send_gs(rs); + else + show_group_stats(rs); + } show_disk_util(); } @@ -717,10 +891,10 @@ static inline void add_stat_sample(struct io_stat *is, unsigned long data) if (data < is->min_val) is->min_val = data; - delta = val - is->mean; + delta = val - is->mean.u.f; if (delta) { - is->mean += delta / (is->samples + 1.0); - is->S += delta * (val - is->mean); + is->mean.u.f += delta / (is->samples + 1.0); + is->S.u.f += delta * (val - is->mean.u.f); } is->samples++; @@ -728,7 +902,7 @@ static inline void add_stat_sample(struct io_stat *is, unsigned long data) static void __add_log_sample(struct io_log *iolog, unsigned long val, enum fio_ddir ddir, unsigned int bs, - unsigned long time) + unsigned long t) { const int nr_samples = iolog->nr_samples; @@ -740,7 +914,7 @@ static void __add_log_sample(struct io_log *iolog, unsigned long val, } iolog->log[nr_samples].val = val; - iolog->log[nr_samples].time = time; + iolog->log[nr_samples].time = t; iolog->log[nr_samples].ddir = ddir; iolog->log[nr_samples].bs = bs; iolog->nr_samples++; @@ -767,6 +941,15 @@ void add_agg_sample(unsigned long val, enum fio_ddir ddir, unsigned int bs) __add_log_sample(iolog, val, ddir, bs, mtime_since_genesis()); } +static void add_clat_percentile_sample(struct thread_stat *ts, + unsigned long usec, enum fio_ddir ddir) +{ + unsigned int idx = plat_val_to_idx(usec); + assert(idx < FIO_IO_U_PLAT_NR); + + ts->io_u_plat[ddir][idx]++; +} + void add_clat_sample(struct thread_data *td, enum fio_ddir ddir, unsigned long usec, unsigned int bs) { @@ -777,8 +960,11 @@ void add_clat_sample(struct thread_data *td, enum fio_ddir ddir, add_stat_sample(&ts->clat_stat[ddir], usec); - if (ts->clat_log) - add_log_sample(td, ts->clat_log, usec, ddir, bs); + if (td->clat_log) + add_log_sample(td, td->clat_log, usec, ddir, bs); + + if (ts->clat_percentiles) + add_clat_percentile_sample(ts, usec, ddir); } void add_slat_sample(struct thread_data *td, enum fio_ddir ddir, @@ -791,8 +977,8 @@ void add_slat_sample(struct thread_data *td, enum fio_ddir ddir, add_stat_sample(&ts->slat_stat[ddir], usec); - if (ts->slat_log) - add_log_sample(td, ts->slat_log, usec, ddir, bs); + if (td->slat_log) + add_log_sample(td, td->slat_log, usec, ddir, bs); } void add_lat_sample(struct thread_data *td, enum fio_ddir ddir, @@ -805,8 +991,8 @@ void add_lat_sample(struct thread_data *td, enum fio_ddir ddir, add_stat_sample(&ts->lat_stat[ddir], usec); - if (ts->lat_log) - add_log_sample(td, ts->lat_log, usec, ddir, bs); + if (td->lat_log) + add_log_sample(td, td->lat_log, usec, ddir, bs); } void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs, @@ -818,17 +1004,43 @@ void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs, if (!ddir_rw(ddir)) return; - spent = mtime_since(&ts->stat_sample_time[ddir], t); + spent = mtime_since(&td->bw_sample_time, t); if (spent < td->o.bw_avg_time) return; - rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) * + rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) * 1000 / spent / 1024; add_stat_sample(&ts->bw_stat[ddir], rate); - if (ts->bw_log) - add_log_sample(td, ts->bw_log, rate, ddir, bs); + if (td->bw_log) + add_log_sample(td, td->bw_log, rate, ddir, bs); + + fio_gettime(&td->bw_sample_time, NULL); + td->stat_io_bytes[ddir] = td->this_io_bytes[ddir]; +} + +void add_iops_sample(struct thread_data *td, enum fio_ddir ddir, + struct timeval *t) +{ + struct thread_stat *ts = &td->ts; + unsigned long spent, iops; + + if (!ddir_rw(ddir)) + return; + + spent = mtime_since(&td->iops_sample_time, t); + if (spent < td->o.iops_avg_time) + return; + + iops = ((td->this_io_blocks[ddir] - td->stat_io_blocks[ddir]) * 1000) / spent; + + add_stat_sample(&ts->iops_stat[ddir], iops); + + if (td->iops_log) { + assert(iops); + add_log_sample(td, td->iops_log, iops, ddir, 0); + } - fio_gettime(&ts->stat_sample_time[ddir], NULL); - ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir]; + fio_gettime(&td->iops_sample_time, NULL); + td->stat_io_blocks[ddir] = td->this_io_blocks[ddir]; }