X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=stat.c;h=a8f2b02665b2b21f278bcef35af984dc2e61e0d0;hp=731b3036544de363e9db4a82084cce1316203d1e;hb=a0a930efb970a688c0e21e396755210e1bdd05a9;hpb=690089990d051d86b4ef2b6fd5c1972c0dd4897b

diff --git a/stat.c b/stat.c
index 731b3036..a8f2b026 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;
+
+/*
+ * Cheasy 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_NETIO | FIO_NULLIO)))
 		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[0], rs->max_run[0]);
+
+		free(p1);
+		free(p2);
+		free(p3);
+		free(p4);
+	}
 }
 
 static void show_disk_util(void)
@@ -331,27 +416,36 @@ static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs,
 			     int ddir)
 {
 	const char *ddir_str[] = { "read ", "write" };
+	struct thread_stat *ts;
 	unsigned long min, max;
 	unsigned long long bw;
 	double mean, dev;
+	char *io_p, *bw_p;
 
 	if (!td->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]);
+	io_p = num2str(td->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, td->runtime[ddir]);
+
+	free(io_p);
+	free(bw_p);
 
-	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);
+	ts = &td->ts;
+	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->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->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(&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, "    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);
 	}
 }
 
@@ -360,33 +454,65 @@ static void show_thread_status(struct thread_data *td,
 {
 	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];
+	int i;
 
-	if (!(td->io_bytes[0] + td->io_bytes[1]) && !td->error)
+	if (!(td->io_bytes[0] + td->io_bytes[1]))
 		return;
 
-	fprintf(f_out, "%s: (groupid=%d): err=%2d:\n",td->name, td->groupid, td->error);
+	if (!td->error)
+		fprintf(f_out, "%s: (groupid=%d): err=%2d: pid=%d\n",td->name, td->groupid, td->error, td->pid);
+	else
+		fprintf(f_out, "%s: (groupid=%d): err=%2d (%s): pid=%d\n",td->name, td->groupid, td->error, td->verror, td->pid);
 
-	show_ddir_status(td, rs, td->ddir);
-	if (td->io_bytes[td->ddir ^ 1])
-		show_ddir_status(td, rs, td->ddir ^ 1);
+	if (td_read(td))
+		show_ddir_status(td, rs, DDIR_READ);
+	if (td_write(td))
+		show_ddir_status(td, rs, DDIR_WRITE);
 
 	runtime = mtime_since(&td->epoch, &td->end_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) td->ts.usr_time * 100 / runt;
+		sys_cpu = (double) td->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, td->ts.ctx);
+
+	/*
+	 * Do depth distribution calculations
+	 */
+	for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
+		io_u_dist[i] = (double) td->io_u_map[i] / (double) td->total_io_u;
+		io_u_dist[i] *= 100.0;
+	}
+
+	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]);
+
+	/*
+	 * Do latency distribution calculations
+	 */
+	for (i = 0; i < FIO_IO_U_LAT_NR; i++) {
+		io_u_lat[i] = (double) td->io_u_lat[i] / (double) td->total_io_u;
+		io_u_lat[i] *= 100.0;
+	}
+
+	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]);
+
+	if (td->description)
+		fprintf(f_out, "%s\n", td->description);
 }
 
 static void show_ddir_status_terse(struct thread_data *td,
 				   struct group_run_stats *rs, int ddir)
 {
+	struct thread_stat *ts = &td->ts;
 	unsigned long min, max;
 	unsigned long long bw;
 	double mean, dev;
@@ -397,17 +523,17 @@ static void show_ddir_status_terse(struct thread_data *td,
 
 	fprintf(f_out, ",%llu,%llu,%lu", td->io_bytes[ddir] >> 10, bw, td->runtime[ddir]);
 
-	if (calc_lat(&td->slat_stat[ddir], &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);
 
-	if (calc_lat(&td->clat_stat[ddir], &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);
 
-	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];
@@ -429,16 +555,16 @@ static void show_thread_status_terse(struct thread_data *td,
 	show_ddir_status_terse(td, rs, 1);
 
 	if (td->runtime[0] + td->runtime[1]) {
-		double runt = td->runtime[0] + td->runtime[1];
+		double runt = (double) (td->runtime[0] + td->runtime[1]);
 
-		usr_cpu = (double) td->usr_time * 100 / runt;
-		sys_cpu = (double) td->sys_time * 100 / runt;
+		usr_cpu = (double) td->ts.usr_time * 100 / runt;
+		sys_cpu = (double) td->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\n", usr_cpu, sys_cpu, td->ts.ctx);
 }
 
 void show_run_stats(void)
@@ -460,11 +586,6 @@ void show_run_stats(void)
 	for_each_td(td, i) {
 		unsigned long long rbw, wbw;
 
-		if (td->error) {
-			fprintf(f_out, "%s: %s\n", td->name, td->verror);
-			continue;
-		}
-
 		rs = &runstats[td->groupid];
 
 		if (td->runtime[0] < rs->min_run[0] || !rs->min_run[0])
@@ -529,20 +650,26 @@ void show_run_stats(void)
 	free(runstats);
 }
 
-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, n;
+
+	if (data > is->max_val)
+		is->max_val = data;
+	if (data < is->min_val)
+		is->min_val = data;
+
+	delta = val - is->mean;
+	n = is->samples + 1.0;
+	is->mean += delta / n;
+	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 +679,62 @@ 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;
 
-	if (td->slat_log)
-		add_log_sample(td, td->slat_log, msec, ddir);
+	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, int ddir, struct timeval *t)
+void add_bw_sample(struct thread_data *td, enum fio_ddir ddir,
+		   struct timeval *t)
 {
-	unsigned long spent = mtime_since(&td->stat_sample_time[ddir], 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];
 }
-
-