X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=stat.c;h=0c2e01b4f3baaf3975ab8c4a5953a7bbc66b60aa;hp=e7eb8ad68aeb369dafe30abb937ad2e5fa1b90bb;hb=f24254e1d142e80dc76d1f2e316da12f6ec125ee;hpb=03b74b3ec5268e731ed7fcaef31c8c0655acd530

diff --git a/stat.c b/stat.c
index e7eb8ad6..0c2e01b4 100644
--- a/stat.c
+++ b/stat.c
@@ -9,274 +9,60 @@
 
 #include "fio.h"
 
-static struct itimerval itimer;
-static struct list_head disk_list = LIST_HEAD_INIT(disk_list);
-
-static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
-{
-	unsigned in_flight;
-	char line[256];
-	FILE *f;
-	char *p;
-
-	f = fopen(du->path, "r");
-	if (!f)
-		return 1;
-
-	p = fgets(line, sizeof(line), f);
-	if (!p) {
-		fclose(f);
-		return 1;
-	}
-
-	if (sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0], &dus->merges[0], &dus->sectors[0], &dus->ticks[0], &dus->ios[1], &dus->merges[1], &dus->sectors[1], &dus->ticks[1], &in_flight, &dus->io_ticks, &dus->time_in_queue) != 11) {
-		fclose(f);
-		return 1;
-	}
-
-	fclose(f);
-	return 0;
-}
-
-static void update_io_tick_disk(struct disk_util *du)
-{
-	struct disk_util_stat __dus, *dus, *ldus;
-	struct timeval t;
-
-	if (get_io_ticks(du, &__dus))
-		return;
-
-	dus = &du->dus;
-	ldus = &du->last_dus;
-
-	dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
-	dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
-	dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
-	dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
-	dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
-	dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
-	dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
-	dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
-	dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
-	dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
-
-	fio_gettime(&t, NULL);
-	du->msec += mtime_since(&du->time, &t);
-	memcpy(&du->time, &t, sizeof(t));
-	memcpy(ldus, &__dus, sizeof(__dus));
-}
-
-void update_io_ticks(void)
-{
-	struct list_head *entry;
-	struct disk_util *du;
-
-	list_for_each(entry, &disk_list) {
-		du = list_entry(entry, struct disk_util, list);
-		update_io_tick_disk(du);
-	}
-}
-
-static int disk_util_exists(dev_t dev)
-{
-	struct list_head *entry;
-	struct disk_util *du;
-
-	list_for_each(entry, &disk_list) {
-		du = list_entry(entry, struct disk_util, list);
-
-		if (du->dev == dev)
-			return 1;
-	}
-
-	return 0;
-}
-
-static void disk_util_add(dev_t dev, char *path)
-{
-	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);
-
-	list_add_tail(&du->list, &disk_list);
-}
-
-static int check_dev_match(dev_t dev, char *path)
-{
-	unsigned int major, minor;
-	char line[256], *p;
-	FILE *f;
-
-	f = fopen(path, "r");
-	if (!f) {
-		perror("open path");
-		return 1;
-	}
-
-	p = fgets(line, sizeof(line), f);
-	if (!p) {
-		fclose(f);
-		return 1;
-	}
-
-	if (sscanf(p, "%u:%u", &major, &minor) != 2) {
-		fclose(f);
-		return 1;
-	}
-
-	if (((major << 8) | minor) == dev) {
-		fclose(f);
-		return 0;
-	}
-
-	fclose(f);
-	return 1;
-}
-
-static int find_block_dir(dev_t dev, char *path)
+/*
+ * Cheesy number->string conversion, complete with carry rounding error.
+ */
+static char *num2str(unsigned long num, int maxlen, int base, int pow2)
 {
-	struct dirent *dir;
-	struct stat st;
-	int found = 0;
-	DIR *D;
+	char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
+	unsigned int thousand;
+	char *buf;
+	int i;
 
-	D = opendir(path);
-	if (!D)
-		return 0;
+	if (pow2)
+		thousand = 1024;
+	else
+		thousand = 1000;
 
-	while ((dir = readdir(D)) != NULL) {
-		char full_path[256];
+	buf = malloc(128);
 
-		if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
-			continue;
+	for (i = 0; base > 1; i++)
+		base /= thousand;
 
-		sprintf(full_path, "%s/%s", path, dir->d_name);
+	do {
+		int len, carry = 0;
 
-		if (!strcmp(dir->d_name, "dev")) {
-			if (!check_dev_match(dev, full_path)) {
-				found = 1;
-				break;
+		len = sprintf(buf, "%'lu", num);
+		if (len <= maxlen) {
+			if (i >= 1) {
+				buf[len] = postfix[i];
+				buf[len + 1] = '\0';
 			}
+			return buf;
 		}
 
-		if (lstat(full_path, &st) == -1) {
-			perror("stat");
-			break;
-		}
+		if ((num % thousand) >= (thousand / 2))
+			carry = 1;
 
-		if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
-			continue;
-
-		found = find_block_dir(dev, full_path);
-		if (found) {
-			strcpy(path, full_path);
-			break;
-		}
-	}
+		num /= thousand;
+		num += carry;
+		i++;
+	} while (i <= 5);
 
-	closedir(D);
-	return found;
-}
-
-void init_disk_util(struct thread_data *td)
-{
-	struct fio_file *f;
-	struct stat st;
-	char foo[PATH_MAX], tmp[PATH_MAX];
-	dev_t dev;
-	char *p;
-
-	if (!td->do_disk_util)
-		return;
-
-	/*
-	 * Just use the same file, they are on the same device.
-	 */
-	f = &td->files[0];
-	if (!stat(f->file_name, &st)) {
-		if (S_ISBLK(st.st_mode))
-			dev = st.st_rdev;
-		else
-			dev = st.st_dev;
-	} else {
-		/*
-		 * must be a file, open "." in that path
-		 */
-		strncpy(foo, f->file_name, PATH_MAX - 1);
-		p = dirname(foo);
-		if (stat(p, &st)) {
-			perror("disk util stat");
-			return;
-		}
-
-		dev = st.st_dev;
-	}
-
-	if (disk_util_exists(dev))
-		return;
-		
-	sprintf(foo, "/sys/block");
-	if (!find_block_dir(dev, foo))
-		return;
-
-	/*
-	 * If there's a ../queue/ directory there, we are inside a partition.
-	 * Check if that is the case and jump back. For loop/md/dm etc we
-	 * are already in the right spot.
-	 */
-	sprintf(tmp, "%s/../queue", foo);
-	if (!stat(tmp, &st)) {
-		p = dirname(foo);
-		sprintf(tmp, "%s/queue", p);
-		if (stat(tmp, &st)) {
-			log_err("unknown sysfs layout\n");
-			return;
-		}
-		strncpy(tmp, p, PATH_MAX - 1);
-		sprintf(foo, "%s", tmp);
-	}
-
-	if (td->ioscheduler)
-		td->sysfs_root = strdup(foo);
-
-	disk_util_add(dev, foo);
-}
-
-void disk_util_timer_arm(void)
-{
-	itimer.it_value.tv_sec = 0;
-	itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
-	setitimer(ITIMER_REAL, &itimer, NULL);
+	return buf;
 }
 
 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,
@@ -303,103 +89,149 @@ static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
 
 static void show_group_stats(struct group_run_stats *rs, int id)
 {
-	fprintf(f_out, "\nRun status group %d (all jobs):\n", id);
+	char *p1, *p2, *p3, *p4;
+	const char *ddir_str[] = { "   READ", "  WRITE" };
+	int i;
 
-	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]);
-}
+	log_info("\nRun status group %d (all jobs):\n", id);
 
-static void show_disk_util(void)
-{
-	struct disk_util_stat *dus;
-	struct list_head *entry, *next;
-	struct disk_util *du;
-	double util;
+	for (i = 0; i <= DDIR_WRITE; i++) {
+		if (!rs->max_run[i])
+			continue;
 
-	fprintf(f_out, "\nDisk stats (read/write):\n");
+		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);
 
-	list_for_each(entry, &disk_list) {
-		du = list_entry(entry, struct disk_util, list);
-		dus = &du->dus;
+		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]);
 
-		util = (double) 100 * du->dus.io_ticks / (double) du->msec;
-		if (util > 100.0)
-			util = 100.0;
+		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;
 
-		fprintf(f_out, "  %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util);
+	/*
+	 * 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;
 
 	/*
-	 * now free the list
+	 * Do latency distribution calculations
 	 */
-	list_for_each_safe(entry, next, &disk_list) {
-		list_del(entry);
-		du = list_entry(entry, struct disk_util, list);
-		free(du->name);
-		free(du);
+	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 thread_data *td, struct group_run_stats *rs,
+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;
+	unsigned long long bw, iops;
 	double mean, dev;
+	char *io_p, *bw_p, *iops_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];
+	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]);
 
-	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);
+	free(iops_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))
+		log_info("    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))
+		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];
-		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);
+		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_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)
+		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);
 
-	show_ddir_status(td, rs, td->ddir);
-	if (td->io_bytes[td->ddir ^ 1])
-		show_ddir_status(td, rs, td->ddir ^ 1);
+	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 = mtime_since(&td->epoch, &td->end_time);
+	runtime = ts->total_run_time;
 	if (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);
+	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_data *td,
+static void show_ddir_status_terse(struct thread_stat *ts,
 				   struct group_run_stats *rs, int ddir)
 {
 	unsigned long min, max;
@@ -407,60 +239,99 @@ 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]);
+	log_info(";%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))
+		log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
 	else
-		fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0);
+		log_info(";%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))
+		log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
 	else
-		fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0);
+		log_info(";%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);
+		log_info(";%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);
-		
+		log_info(";%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);
+	log_info("%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 = (double) (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);
+	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;
-	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));
 
@@ -472,42 +343,127 @@ 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->error) {
-			fprintf(f_out, "%s: %s\n", td->name, td->verror);
+		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;
 		}
 
-		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;
+		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++) {
@@ -525,13 +481,14 @@ 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) {
@@ -542,12 +499,13 @@ void show_run_stats(void)
 	}
 
 	free(runstats);
+	free(threadstats);
 }
 
 static inline void add_stat_sample(struct io_stat *is, unsigned long data)
 {
 	double val = data;
-	double delta, n;
+	double delta;
 
 	if (data > is->max_val)
 		is->max_val = data;
@@ -555,15 +513,16 @@ static inline void add_stat_sample(struct io_stat *is, unsigned long data)
 		is->min_val = data;
 
 	delta = val - is->mean;
-	n = is->samples + 1.0;
-	is->mean += delta / n;
-	is->S += 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 thread_data *td, struct io_log *iolog,
-			   unsigned long val, enum fio_ddir 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;
@@ -573,46 +532,62 @@ 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++;
 }
 
+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)
 {
-	add_stat_sample(&td->clat_stat[ddir], msec);
+	struct thread_stat *ts = &td->ts;
+
+	add_stat_sample(&ts->clat_stat[ddir], msec);
 
-	if (td->clat_log)
-		add_log_sample(td, td->clat_log, msec, ddir);
+	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)
 {
-	add_stat_sample(&td->slat_stat[ddir], msec);
+	struct thread_stat *ts = &td->ts;
+
+	add_stat_sample(&ts->slat_stat[ddir], msec);
 
-	if (td->slat_log)
-		add_log_sample(td, td->slat_log, msec, ddir);
+	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)
 {
-	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)
+	if (spent < td->o.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];
 }
-
-