[fio.git] / stat.c

#include <stdio.h>
#include <string.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <dirent.h>
#include <libgen.h>
#include <math.h>

#include "fio.h"

/*
 * Cheesy number->string conversion, complete with carry rounding error.
 */
static char *num2str(unsigned long num, int maxlen, int base, int pow2)
{
	char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
	unsigned int thousand;
	char *buf;
	int i;

	if (pow2)
		thousand = 1024;
	else
		thousand = 1000;

	buf = malloc(128);

	for (i = 0; base > 1; i++)
		base /= thousand;

	do {
		int len, carry = 0;

		len = sprintf(buf, "%'lu", num);
		if (len <= maxlen) {
			if (i >= 1) {
				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;
}

void update_rusage_stat(struct thread_data *td)
{
	struct thread_stat *ts = &td->ts;

	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(&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 = is->samples;

	if (is->samples == 0)
		return 0;

	*min = is->min_val;
	*max = is->max_val;

	n = (double) is->samples;
	*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;

	log_info("\nRun status group %d (all jobs):\n", id);

	for (i = 0; i <= DDIR_WRITE; i++) {
		if (!rs->max_run[i])
			continue;

		p1 = num2str(rs->io_kb[i], 6, 1000, 1);
		p2 = num2str(rs->agg[i], 6, 1000, 1);
		p3 = num2str(rs->min_bw[i], 6, 1000, 1);
		p4 = num2str(rs->max_bw[i], 6, 1000, 1);

		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]);

		free(p1);
		free(p2);
		free(p3);
		free(p4);
	}
}

#define ts_total_io_u(ts)	\
	((ts)->total_io_u[0] + (ts)->total_io_u[1])

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(ts);
		io_u_dist[i] *= 100.0;
		if (io_u_dist[i] < 0.1 && ts->io_u_map[i])
			io_u_dist[i] = 0.1;
	}
}

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(ts);
		io_u_lat[i] *= 100.0;
		if (io_u_lat[i] < 0.01 && ts->io_u_lat[i])
			io_u_lat[i] = 0.01;
	}
}

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, iops;
	double mean, dev;
	char *io_p, *bw_p, *iops_p;

	if (!ts->runtime[ddir])
		return;

	bw = ts->io_bytes[ddir] / ts->runtime[ddir];
	iops = (1000 * ts->total_io_u[ddir]) / ts->runtime[ddir];
	io_p = num2str(ts->io_bytes[ddir] >> 10, 6, 1000, 1);
	bw_p = num2str(bw, 6, 1000, 1);
	iops_p = num2str(iops, 6, 1, 0);

	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]);

	free(io_p);
	free(bw_p);
	free(iops_p);

	if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
		log_info("    slat (msec): min=%5lu, max=%5lu, avg=%5.02f, stdev=%5.02f\n", min, max, mean, dev);

	if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
		log_info("    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];
		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);
	}
}

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 (!(ts->io_bytes[0] + ts->io_bytes[1]))
		return;

	if (!ts->error)
		log_info("%s: (groupid=%d, jobs=%d): err=%2d: pid=%d\n", ts->name, ts->groupid, ts->members, ts->error, ts->pid);
	else
		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);

	if (ts->description)
		log_info("  Description  : [%s]\n", ts->description);

	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 = (double) runtime;

		usr_cpu = (double) ts->usr_time * 100 / runt;
		sys_cpu = (double) ts->sys_time * 100 / runt;
	} else {
		usr_cpu = 0;
		sys_cpu = 0;
	}

	log_info("  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);

	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]);
	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]);

	log_info("     lat (msec): 2=%3.2f%%, 4=%3.2f%%, 10=%3.2f%%, 20=%3.2f%%, 50=%3.2f%%, 100=%3.2f%%\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]);
	log_info("     lat (msec): 250=%3.2f%%, 500=%3.2f%%, 750=%3.2f%%, 1000=%3.2f%%, >=2000=%3.2f%%\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_stat *ts,
				   struct group_run_stats *rs, int ddir)
{
	unsigned long min, max;
	unsigned long long bw;
	double mean, dev;

	bw = 0;
	if (ts->runtime[ddir])
		bw = ts->io_bytes[ddir] / ts->runtime[ddir];

	log_info(";%llu;%llu;%lu", ts->io_bytes[ddir] >> 10, bw, ts->runtime[ddir]);

	if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
		log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
	else
		log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);

	if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
		log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
	else
		log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);

	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(";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
	} else
		log_info(";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0);
}


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;

	log_info("%s;%d;%d", ts->name, ts->groupid, ts->error);

	show_ddir_status_terse(ts, rs, 0);
	show_ddir_status_terse(ts, rs, 1);

	if (ts->total_run_time) {
		double runt = (double) ts->total_run_time;

		usr_cpu = (double) ts->usr_time * 100 / runt;
		sys_cpu = (double) ts->sys_time * 100 / runt;
	} else {
		usr_cpu = 0;
		sys_cpu = 0;
	}

	log_info(";%f%%;%f%%;%lu", usr_cpu, sys_cpu, ts->ctx);

	stat_calc_dist(ts, io_u_dist);
	stat_calc_lat(ts, io_u_lat);

	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]);

	log_info(";%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%\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]);
	log_info(";%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%", io_u_lat[6], io_u_lat[7], io_u_lat[8], io_u_lat[9], io_u_lat[10]);

	if (ts->description)
		log_info(";%s", ts->description);

	log_info("\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;
	struct thread_stat *threadstats, *ts;
	int i, j, k, l, nr_ts, last_ts, idx;

	runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));

	for (i = 0; i < groupid + 1; i++) {
		rs = &runstats[i];

		memset(rs, 0, sizeof(*rs));
		rs->min_bw[0] = rs->min_run[0] = ~0UL;
		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) {
		if (!td->o.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 <= DDIR_WRITE; j++) {
			ts->clat_stat[j].min_val = -1UL;
			ts->slat_stat[j].min_val = -1UL;
			ts->bw_stat[j].min_val = -1UL;
		}
		ts->groupid = -1;
	}

	j = 0;
	last_ts = -1;
	idx = 0;
	for_each_td(td, i) {
		if (idx && (!td->o.group_reporting ||
		    (td->o.group_reporting && last_ts != td->groupid))) {
			idx = 0;
			j++;
		}

		last_ts = td->groupid;

		ts = &threadstats[j];

		idx++;
		ts->members++;

		if (ts->groupid == -1) {
			/*
			 * These are per-group shared already
			 */
			ts->name = td->o.name;
			ts->description = td->o.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;
		}

		for (l = 0; l <= DDIR_WRITE; 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];

		for (k = 0; k <= DDIR_WRITE; k++) {
			ts->total_io_u[k] += td->ts.total_io_u[k];
			ts->short_io_u[k] += td->ts.short_io_u[k];
		}

		ts->total_run_time += td->ts.total_run_time;
	}

	for (i = 0; i < nr_ts; i++) {
		unsigned long long bw;

		ts = &threadstats[i];
		rs = &runstats[ts->groupid];

		for (j = 0; j <= DDIR_WRITE; 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++) {
		rs = &runstats[i];

		if (rs->max_run[0])
			rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0];
		if (rs->max_run[1])
			rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1];
	}

	/*
	 * don't overwrite last signal output
	 */
	if (!terse_output)
		printf("\n");

	for (i = 0; i < nr_ts; i++) {
		ts = &threadstats[i];
		rs = &runstats[ts->groupid];

		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);

		show_disk_util();
	}

	free(runstats);
	free(threadstats);
}

static inline void add_stat_sample(struct io_stat *is, unsigned long data)
{
	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;
	if (delta) {
		is->mean += delta / (is->samples + 1.0);
		is->S += delta * (val - is->mean);
	}

	is->samples++;
}

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;

		iolog->log = realloc(iolog->log, new_size);
		iolog->max_samples <<= 1;
	}

	iolog->log[iolog->nr_samples].val = val;
	iolog->log[iolog->nr_samples].time = time;
	iolog->log[iolog->nr_samples].ddir = ddir;
	iolog->nr_samples++;
}

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)
{
	struct io_log *iolog = agg_io_log[ddir];

	__add_log_sample(iolog, val, ddir, mtime_since_genesis());
}

void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
		     unsigned long msec)
{
	struct thread_stat *ts = &td->ts;

	add_stat_sample(&ts->clat_stat[ddir], msec);

	if (ts->clat_log)
		add_log_sample(td, ts->clat_log, msec, ddir);
}

void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
		     unsigned long msec)
{
	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->o.bw_avg_time)
		return;

	rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) / spent;
	add_stat_sample(&ts->bw_stat[ddir], rate);

	if (ts->bw_log)
		add_log_sample(td, ts->bw_log, rate, ddir);

	fio_gettime(&ts->stat_sample_time[ddir], NULL);
	ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
}
Commit	Line	Data
3c39a379 JA	1	#include <stdio.h>
	2	#include <string.h>
	3	#include <sys/time.h>
	4	#include <sys/types.h>
5c4e1dbc	5	#include <sys/stat.h>
3c39a379 JA	6	#include <dirent.h>
	7	#include <libgen.h>
	8	#include <math.h>
	9
	10	#include "fio.h"
3c39a379	11
dbe1125e	12	/*
34403fb1	13	* Cheesy number->string conversion, complete with carry rounding error.
dbe1125e	14	*/
b3605062	15	static char *num2str(unsigned long num, int maxlen, int base, int pow2)
dbe1125e	16	{
b3605062 JA	17	char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
b3605062 JA	18	unsigned int thousand;
dbe1125e JA	19	char *buf;
	20	int i;
	21
b3605062 JA	22	if (pow2)
	23	thousand = 1024;
	24	else
	25	thousand = 1000;
	26
dbe1125e JA	27	buf = malloc(128);
	28
	29	for (i = 0; base > 1; i++)
	30	base /= thousand;
	31
	32	do {
	33	int len, carry = 0;
	34
	35	len = sprintf(buf, "%'lu", num);
	36	if (len <= maxlen) {
b3605062 JA	37	if (i >= 1) {
	38	buf[len] = postfix[i];
	39	buf[len + 1] = '\0';
	40	}
dbe1125e JA	41	return buf;
	42	}
	43
	44	if ((num % thousand) >= (thousand / 2))
	45	carry = 1;
	46
	47	num /= thousand;
	48	num += carry;
	49	i++;
f3502ba2	50	} while (i <= 5);
dbe1125e JA	51
	52	return buf;
	53	}
	54
3c39a379 JA	55	void update_rusage_stat(struct thread_data *td)
3c39a379 JA	56	{
756867bd	57	struct thread_stat *ts = &td->ts;
3c39a379	58
079ad09b JA	59	getrusage(RUSAGE_SELF, &ts->ru_end);
	60
	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);
3c39a379	64
079ad09b	65	memcpy(&ts->ru_start, &ts->ru_end, sizeof(ts->ru_end));
3c39a379 JA	66	}
	67
	68	static int calc_lat(struct io_stat is, unsigned long min, unsigned long *max,
	69	double mean, double dev)
	70	{
68704084	71	double n = is->samples;
3c39a379 JA	72
	73	if (is->samples == 0)
	74	return 0;
	75
	76	*min = is->min_val;
	77	*max = is->max_val;
	78
	79	n = (double) is->samples;
68704084	80	*mean = is->mean;
e6d276f2	81
68704084 JA	82	if (n > 1.0)
68704084 JA	83	*dev = sqrt(is->S / (n - 1.0));
ef9c5c40	84	else
68704084	85	*dev = -1.0;
ef9c5c40	86
3c39a379 JA	87	return 1;
	88	}
	89
	90	static void show_group_stats(struct group_run_stats *rs, int id)
	91	{
dbe1125e JA	92	char p1, p2, p3, p4;
	93	const char *ddir_str[] = { " READ", " WRITE" };
	94	int i;
	95
6d86144d	96	log_info("\nRun status group %d (all jobs):\n", id);
3c39a379	97
dbe1125e JA	98	for (i = 0; i <= DDIR_WRITE; i++) {
	99	if (!rs->max_run[i])
	100	continue;
	101
b3605062 JA	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);
dbe1125e	106
6d86144d	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]);
dbe1125e JA	108
	109	free(p1);
	110	free(p2);
	111	free(p3);
	112	free(p4);
	113	}
3c39a379 JA	114	}
3c39a379 JA	115
b3605062 JA	116	#define ts_total_io_u(ts) \
	117	((ts)->total_io_u[0] + (ts)->total_io_u[1])
	118
2270890c JA	119	static void stat_calc_dist(struct thread_stat ts, double io_u_dist)
	120	{
	121	int i;
	122
	123	/*
	124	* Do depth distribution calculations
	125	*/
	126	for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
b3605062	127	io_u_dist[i] = (double) ts->io_u_map[i] / (double) ts_total_io_u(ts);
2270890c	128	io_u_dist[i] *= 100.0;
d0f62ba3 JA	129	if (io_u_dist[i] < 0.1 && ts->io_u_map[i])
d0f62ba3 JA	130	io_u_dist[i] = 0.1;
2270890c JA	131	}
	132	}
	133
	134	static void stat_calc_lat(struct thread_stat ts, double io_u_lat)
	135	{
	136	int i;
	137
	138	/*
	139	* Do latency distribution calculations
	140	*/
	141	for (i = 0; i < FIO_IO_U_LAT_NR; i++) {
b3605062	142	io_u_lat[i] = (double) ts->io_u_lat[i] / (double) ts_total_io_u(ts);
2270890c	143	io_u_lat[i] *= 100.0;
d0f62ba3 JA	144	if (io_u_lat[i] < 0.01 && ts->io_u_lat[i])
d0f62ba3 JA	145	io_u_lat[i] = 0.01;
2270890c JA	146	}
	147	}
	148
756867bd	149	static void show_ddir_status(struct group_run_stats rs, struct thread_stat ts,
3c39a379 JA	150	int ddir)
3c39a379 JA	151	{
3c9b60c1	152	const char *ddir_str[] = { "read ", "write" };
3c39a379	153	unsigned long min, max;
b3605062	154	unsigned long long bw, iops;
3c39a379	155	double mean, dev;
b3605062	156	char io_p, bw_p, *iops_p;
3c39a379	157
756867bd	158	if (!ts->runtime[ddir])
3c39a379 JA	159	return;
3c39a379 JA	160
756867bd	161	bw = ts->io_bytes[ddir] / ts->runtime[ddir];
b3605062 JA	162	iops = (1000 * ts->total_io_u[ddir]) / ts->runtime[ddir];
	163	io_p = num2str(ts->io_bytes[ddir] >> 10, 6, 1000, 1);
	164	bw_p = num2str(bw, 6, 1000, 1);
	165	iops_p = num2str(iops, 6, 1, 0);
dbe1125e	166
6d86144d	167	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]);
dbe1125e JA	168
	169	free(io_p);
	170	free(bw_p);
b3605062	171	free(iops_p);
3c39a379	172
079ad09b	173	if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
6d86144d	174	log_info(" slat (msec): min=%5lu, max=%5lu, avg=%5.02f, stdev=%5.02f\n", min, max, mean, dev);
3c39a379	175
079ad09b	176	if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
6d86144d	177	log_info(" clat (msec): min=%5lu, max=%5lu, avg=%5.02f, stdev=%5.02f\n", min, max, mean, dev);
3c39a379	178
079ad09b	179	if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
3c39a379 JA	180	double p_of_agg;
	181
	182	p_of_agg = mean * 100 / (double) rs->agg[ddir];
6d86144d	183	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);
3c39a379 JA	184	}
	185	}
	186
756867bd	187	static void show_thread_status(struct thread_stat *ts,
3c39a379 JA	188	struct group_run_stats *rs)
	189	{
	190	double usr_cpu, sys_cpu;
69008999	191	unsigned long runtime;
71619dc2	192	double io_u_dist[FIO_IO_U_MAP_NR];
ec118304	193	double io_u_lat[FIO_IO_U_LAT_NR];
3c39a379	194
756867bd	195	if (!(ts->io_bytes[0] + ts->io_bytes[1]))
3c39a379 JA	196	return;
3c39a379 JA	197
756867bd	198	if (!ts->error)
6d86144d	199	log_info("%s: (groupid=%d, jobs=%d): err=%2d: pid=%d\n", ts->name, ts->groupid, ts->members, ts->error, ts->pid);
6d663077	200	else
6d86144d	201	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);
3c39a379	202
7bdce1bd	203	if (ts->description)
6d86144d	204	log_info(" Description : [%s]\n", ts->description);
7bdce1bd	205
756867bd JA	206	if (ts->io_bytes[DDIR_READ])
	207	show_ddir_status(rs, ts, DDIR_READ);
	208	if (ts->io_bytes[DDIR_WRITE])
	209	show_ddir_status(rs, ts, DDIR_WRITE);
3c39a379	210
756867bd	211	runtime = ts->total_run_time;
69008999	212	if (runtime) {
1e97cce9	213	double runt = (double) runtime;
3c39a379	214
756867bd JA	215	usr_cpu = (double) ts->usr_time * 100 / runt;
756867bd JA	216	sys_cpu = (double) ts->sys_time * 100 / runt;
3c39a379 JA	217	} else {
	218	usr_cpu = 0;
	219	sys_cpu = 0;
	220	}
	221
6d86144d	222	log_info(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, ts->ctx);
71619dc2	223
2270890c JA	224	stat_calc_dist(ts, io_u_dist);
2270890c JA	225	stat_calc_lat(ts, io_u_lat);
71619dc2	226
6d86144d	227	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]);
30061b97	228	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]);
61697c37	229
d0f62ba3 JA	230	log_info(" lat (msec): 2=%3.2f%%, 4=%3.2f%%, 10=%3.2f%%, 20=%3.2f%%, 50=%3.2f%%, 100=%3.2f%%\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]);
d0f62ba3 JA	231	log_info(" lat (msec): 250=%3.2f%%, 500=%3.2f%%, 750=%3.2f%%, 1000=%3.2f%%, >=2000=%3.2f%%\n", io_u_lat[6], io_u_lat[7], io_u_lat[8], io_u_lat[9], io_u_lat[10]);
3c39a379 JA	232	}
3c39a379 JA	233
756867bd	234	static void show_ddir_status_terse(struct thread_stat *ts,
c6ae0a5b JA	235	struct group_run_stats *rs, int ddir)
	236	{
	237	unsigned long min, max;
	238	unsigned long long bw;
	239	double mean, dev;
	240
	241	bw = 0;
756867bd JA	242	if (ts->runtime[ddir])
756867bd JA	243	bw = ts->io_bytes[ddir] / ts->runtime[ddir];
c6ae0a5b	244
6d86144d	245	log_info(";%llu;%llu;%lu", ts->io_bytes[ddir] >> 10, bw, ts->runtime[ddir]);
c6ae0a5b	246
079ad09b	247	if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
6d86144d	248	log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
c6ae0a5b	249	else
6d86144d	250	log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
c6ae0a5b	251
079ad09b	252	if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
6d86144d	253	log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
c6ae0a5b	254	else
6d86144d	255	log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
c6ae0a5b	256
079ad09b	257	if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
c6ae0a5b JA	258	double p_of_agg;
	259
	260	p_of_agg = mean * 100 / (double) rs->agg[ddir];
6d86144d	261	log_info(";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
c6ae0a5b	262	} else
6d86144d	263	log_info(";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0);
c6ae0a5b JA	264	}
	265
	266
756867bd	267	static void show_thread_status_terse(struct thread_stat *ts,
c6ae0a5b JA	268	struct group_run_stats *rs)
c6ae0a5b JA	269	{
2270890c JA	270	double io_u_dist[FIO_IO_U_MAP_NR];
2270890c JA	271	double io_u_lat[FIO_IO_U_LAT_NR];
c6ae0a5b JA	272	double usr_cpu, sys_cpu;
c6ae0a5b JA	273
6d86144d	274	log_info("%s;%d;%d", ts->name, ts->groupid, ts->error);
c6ae0a5b	275
756867bd JA	276	show_ddir_status_terse(ts, rs, 0);
756867bd JA	277	show_ddir_status_terse(ts, rs, 1);
c6ae0a5b	278
756867bd JA	279	if (ts->total_run_time) {
756867bd JA	280	double runt = (double) ts->total_run_time;
c6ae0a5b	281
756867bd JA	282	usr_cpu = (double) ts->usr_time * 100 / runt;
756867bd JA	283	sys_cpu = (double) ts->sys_time * 100 / runt;
c6ae0a5b JA	284	} else {
	285	usr_cpu = 0;
	286	sys_cpu = 0;
	287	}
	288
6d86144d	289	log_info(";%f%%;%f%%;%lu", usr_cpu, sys_cpu, ts->ctx);
2270890c JA	290
	291	stat_calc_dist(ts, io_u_dist);
	292	stat_calc_lat(ts, io_u_lat);
	293
6d86144d	294	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]);
2270890c	295
d0f62ba3 JA	296	log_info(";%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%\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]);
d0f62ba3 JA	297	log_info(";%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%;%3.2f%%", io_u_lat[6], io_u_lat[7], io_u_lat[8], io_u_lat[9], io_u_lat[10]);
2270890c JA	298
2270890c JA	299	if (ts->description)
6d86144d	300	log_info(";%s", ts->description);
2270890c	301
6d86144d	302	log_info("\n");
756867bd JA	303	}
756867bd JA	304
197574e4	305	static void sum_stat(struct io_stat dst, struct io_stat src, int nr)
756867bd JA	306	{
	307	double mean, S;
	308
	309	dst->min_val = min(dst->min_val, src->min_val);
	310	dst->max_val = max(dst->max_val, src->max_val);
	311	dst->samples += src->samples;
	312
	313	/*
	314	* Needs a new method for calculating stddev, we cannot just
	315	* average them we do below for nr > 1
	316	*/
	317	if (nr == 1) {
	318	mean = src->mean;
	319	S = src->S;
	320	} else {
c39cced6 JA	321	mean = ((src->mean * (double) (nr - 1)) + dst->mean) / ((double) nr);
c39cced6 JA	322	S = ((src->S * (double) (nr - 1)) + dst->S) / ((double) nr);
756867bd JA	323	}
	324
	325	dst->mean = mean;
	326	dst->S = S;
	327	}
	328
3c39a379 JA	329	void show_run_stats(void)
	330	{
	331	struct group_run_stats runstats, rs;
	332	struct thread_data *td;
756867bd	333	struct thread_stat threadstats, ts;
197574e4	334	int i, j, k, l, nr_ts, last_ts, idx;
3c39a379 JA	335
	336	runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
	337
	338	for (i = 0; i < groupid + 1; i++) {
	339	rs = &runstats[i];
	340
	341	memset(rs, 0, sizeof(*rs));
	342	rs->min_bw[0] = rs->min_run[0] = ~0UL;
	343	rs->min_bw[1] = rs->min_run[1] = ~0UL;
	344	}
	345
756867bd JA	346	/*
	347	* find out how many threads stats we need. if group reporting isn't
	348	* enabled, it's one-per-td.
	349	*/
	350	nr_ts = 0;
	351	last_ts = -1;
	352	for_each_td(td, i) {
2dc1bbeb	353	if (!td->o.group_reporting) {
756867bd JA	354	nr_ts++;
	355	continue;
	356	}
	357	if (last_ts == td->groupid)
	358	continue;
	359
	360	last_ts = td->groupid;
	361	nr_ts++;
	362	}
	363
	364	threadstats = malloc(nr_ts * sizeof(struct thread_stat));
	365
	366	for (i = 0; i < nr_ts; i++) {
	367	ts = &threadstats[i];
	368
	369	memset(ts, 0, sizeof(*ts));
de64df05	370	for (j = 0; j <= DDIR_WRITE; j++) {
197574e4 JA	371	ts->clat_stat[j].min_val = -1UL;
	372	ts->slat_stat[j].min_val = -1UL;
	373	ts->bw_stat[j].min_val = -1UL;
	374	}
7abd0e3a	375	ts->groupid = -1;
756867bd JA	376	}
	377
	378	j = 0;
	379	last_ts = -1;
197574e4	380	idx = 0;
34572e28	381	for_each_td(td, i) {
2dc1bbeb JA	382	if (idx && (!td->o.group_reporting \|\|
2dc1bbeb JA	383	(td->o.group_reporting && last_ts != td->groupid))) {
7abd0e3a JA	384	idx = 0;
	385	j++;
	386	}
	387
	388	last_ts = td->groupid;
	389
756867bd JA	390	ts = &threadstats[j];
756867bd JA	391
197574e4	392	idx++;
6586ee89	393	ts->members++;
756867bd	394
7abd0e3a	395	if (ts->groupid == -1) {
2dc84ba7 JA	396	/*
	397	* These are per-group shared already
	398	*/
2dc1bbeb JA	399	ts->name = td->o.name;
2dc1bbeb JA	400	ts->description = td->o.description;
756867bd	401	ts->groupid = td->groupid;
2dc84ba7 JA	402
	403	/*
	404	* first pid in group, not very useful...
	405	*/
756867bd	406	ts->pid = td->pid;
2dc84ba7 JA	407	}
	408
	409	if (td->error && !ts->error) {
	410	ts->error = td->error;
756867bd JA	411	ts->verror = td->verror;
	412	}
	413
de64df05	414	for (l = 0; l <= DDIR_WRITE; l++) {
197574e4 JA	415	sum_stat(&ts->clat_stat[l], &td->ts.clat_stat[l], idx);
	416	sum_stat(&ts->slat_stat[l], &td->ts.slat_stat[l], idx);
	417	sum_stat(&ts->bw_stat[l], &td->ts.bw_stat[l], idx);
756867bd	418
197574e4 JA	419	ts->stat_io_bytes[l] += td->ts.stat_io_bytes[l];
	420	ts->io_bytes[l] += td->ts.io_bytes[l];
	421
	422	if (ts->runtime[l] < td->ts.runtime[l])
	423	ts->runtime[l] = td->ts.runtime[l];
	424	}
756867bd JA	425
	426	ts->usr_time += td->ts.usr_time;
	427	ts->sys_time += td->ts.sys_time;
	428	ts->ctx += td->ts.ctx;
	429
	430	for (k = 0; k < FIO_IO_U_MAP_NR; k++)
	431	ts->io_u_map[k] += td->ts.io_u_map[k];
	432	for (k = 0; k < FIO_IO_U_LAT_NR; k++)
	433	ts->io_u_lat[k] += td->ts.io_u_lat[k];
	434
30061b97	435	for (k = 0; k <= DDIR_WRITE; k++) {
b3605062	436	ts->total_io_u[k] += td->ts.total_io_u[k];
30061b97 JA	437	ts->short_io_u[k] += td->ts.short_io_u[k];
30061b97 JA	438	}
756867bd JA	439
756867bd JA	440	ts->total_run_time += td->ts.total_run_time;
756867bd JA	441	}
	442
	443	for (i = 0; i < nr_ts; i++) {
94370ac4	444	unsigned long long bw;
3c39a379	445
756867bd JA	446	ts = &threadstats[i];
756867bd JA	447	rs = &runstats[ts->groupid];
3c39a379	448
de64df05	449	for (j = 0; j <= DDIR_WRITE; j++) {
94370ac4 JA	450	if (!ts->runtime[j])
	451	continue;
	452	if (ts->runtime[j] < rs->min_run[j] \|\| !rs->min_run[j])
	453	rs->min_run[j] = ts->runtime[j];
	454	if (ts->runtime[j] > rs->max_run[j])
	455	rs->max_run[j] = ts->runtime[j];
	456
	457	bw = 0;
	458	if (ts->runtime[j])
	459	bw = ts->io_bytes[j] / (unsigned long long) ts->runtime[j];
	460	if (bw < rs->min_bw[j])
	461	rs->min_bw[j] = bw;
	462	if (bw > rs->max_bw[j])
	463	rs->max_bw[j] = bw;
	464
	465	rs->io_kb[j] += ts->io_bytes[j] >> 10;
	466	}
3c39a379 JA	467	}
	468
	469	for (i = 0; i < groupid + 1; i++) {
	470	rs = &runstats[i];
	471
	472	if (rs->max_run[0])
	473	rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0];
	474	if (rs->max_run[1])
	475	rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1];
	476	}
	477
	478	/*
	479	* don't overwrite last signal output
	480	*/
c6ae0a5b JA	481	if (!terse_output)
c6ae0a5b JA	482	printf("\n");
3c39a379	483
756867bd JA	484	for (i = 0; i < nr_ts; i++) {
	485	ts = &threadstats[i];
	486	rs = &runstats[ts->groupid];
3c39a379	487
c6ae0a5b	488	if (terse_output)
756867bd	489	show_thread_status_terse(ts, rs);
c6ae0a5b	490	else
756867bd	491	show_thread_status(ts, rs);
3c39a379 JA	492	}
3c39a379 JA	493
c6ae0a5b JA	494	if (!terse_output) {
	495	for (i = 0; i < groupid + 1; i++)
	496	show_group_stats(&runstats[i], i);
3c39a379	497
c6ae0a5b JA	498	show_disk_util();
c6ae0a5b JA	499	}
eecf272f JA	500
eecf272f JA	501	free(runstats);
756867bd	502	free(threadstats);
3c39a379 JA	503	}
3c39a379 JA	504
68704084	505	static inline void add_stat_sample(struct io_stat *is, unsigned long data)
3c39a379	506	{
68704084	507	double val = data;
6660cc67	508	double delta;
68704084 JA	509
	510	if (data > is->max_val)
	511	is->max_val = data;
	512	if (data < is->min_val)
	513	is->min_val = data;
	514
	515	delta = val - is->mean;
ef11d737 JA	516	if (delta) {
	517	is->mean += delta / (is->samples + 1.0);
	518	is->S += delta * (val - is->mean);
	519	}
3c39a379	520
3c39a379 JA	521	is->samples++;
	522	}
	523
bb3884d8 JA	524	static void __add_log_sample(struct io_log *iolog, unsigned long val,
bb3884d8 JA	525	enum fio_ddir ddir, unsigned long time)
3c39a379 JA	526	{
	527	if (iolog->nr_samples == iolog->max_samples) {
	528	int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
	529
	530	iolog->log = realloc(iolog->log, new_size);
	531	iolog->max_samples <<= 1;
	532	}
	533
	534	iolog->log[iolog->nr_samples].val = val;
bb3884d8	535	iolog->log[iolog->nr_samples].time = time;
3c39a379 JA	536	iolog->log[iolog->nr_samples].ddir = ddir;
	537	iolog->nr_samples++;
	538	}
	539
bb3884d8 JA	540	static void add_log_sample(struct thread_data td, struct io_log iolog,
	541	unsigned long val, enum fio_ddir ddir)
	542	{
	543	__add_log_sample(iolog, val, ddir, mtime_since_now(&td->epoch));
	544	}
	545
	546	void add_agg_sample(unsigned long val, enum fio_ddir ddir)
	547	{
	548	struct io_log *iolog = agg_io_log[ddir];
	549
	550	__add_log_sample(iolog, val, ddir, mtime_since_genesis());
	551	}
	552
1e97cce9 JA	553	void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
1e97cce9 JA	554	unsigned long msec)
3c39a379	555	{
756867bd	556	struct thread_stat *ts = &td->ts;
079ad09b JA	557
079ad09b JA	558	add_stat_sample(&ts->clat_stat[ddir], msec);
3c39a379	559
079ad09b JA	560	if (ts->clat_log)
079ad09b JA	561	add_log_sample(td, ts->clat_log, msec, ddir);
3c39a379 JA	562	}
3c39a379 JA	563
1e97cce9 JA	564	void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
1e97cce9 JA	565	unsigned long msec)
3c39a379	566	{
756867bd	567	struct thread_stat *ts = &td->ts;
079ad09b JA	568
079ad09b JA	569	add_stat_sample(&ts->slat_stat[ddir], msec);
3c39a379	570
079ad09b JA	571	if (ts->slat_log)
079ad09b JA	572	add_log_sample(td, ts->slat_log, msec, ddir);
3c39a379 JA	573	}
3c39a379 JA	574
1e97cce9 JA	575	void add_bw_sample(struct thread_data *td, enum fio_ddir ddir,
1e97cce9 JA	576	struct timeval *t)
3c39a379	577	{
756867bd	578	struct thread_stat *ts = &td->ts;
079ad09b	579	unsigned long spent = mtime_since(&ts->stat_sample_time[ddir], t);
3c39a379 JA	580	unsigned long rate;
3c39a379 JA	581
2dc1bbeb	582	if (spent < td->o.bw_avg_time)
3c39a379 JA	583	return;
3c39a379 JA	584
079ad09b JA	585	rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) / spent;
079ad09b JA	586	add_stat_sample(&ts->bw_stat[ddir], rate);
3c39a379	587
079ad09b JA	588	if (ts->bw_log)
079ad09b JA	589	add_log_sample(td, ts->bw_log, rate, ddir);
3c39a379	590
079ad09b JA	591	fio_gettime(&ts->stat_sample_time[ddir], NULL);
079ad09b JA	592	ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
3c39a379	593	}