X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=stat.c;h=4fe0dbd3dd65a9a129c9a6951bfe9e9bdae3754a;hp=f9c6985338d998095fd35a229d8b0be5ed533760;hb=148cf6a95f738d6d9afd62bc63ba40b64bd647bd;hpb=f3502ba2dd72d846a388c774a01c6bb833871248

diff --git a/stat.c b/stat.c
index f9c69853..4fe0dbd3 100644
--- a/stat.c
+++ b/stat.c
@@ -11,21 +11,23 @@
 
 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.
+ * Cheesy number->string conversion, complete with carry rounding error.
  */
-static char *num2str(unsigned long num, int maxlen, int base)
+static char *num2str(unsigned long num, int maxlen, int base, int pow2)
 {
-	/*
-	 * 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 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++)
@@ -36,8 +38,10 @@ static char *num2str(unsigned long num, int maxlen, int base)
 
 		len = sprintf(buf, "%'lu", num);
 		if (len <= maxlen) {
-			buf[len] = postfix[i];
-			buf[len + 1] = '\0';
+			if (i >= 1) {
+				buf[len] = postfix[i];
+				buf[len + 1] = '\0';
+			}
 			return buf;
 		}
 
@@ -237,21 +241,13 @@ static int find_block_dir(dev_t dev, char *path)
 	return found;
 }
 
-void init_disk_util(struct thread_data *td)
+static void __init_disk_util(struct thread_data *td, struct fio_file *f)
 {
-	struct fio_file *f;
 	struct stat st;
 	char foo[PATH_MAX], tmp[PATH_MAX];
 	dev_t dev;
 	char *p;
 
-	if (!td->do_disk_util || (td->io_ops->flags & (FIO_NETIO | FIO_NULLIO)))
-		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;
@@ -273,6 +269,17 @@ 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))
@@ -295,12 +302,25 @@ void init_disk_util(struct thread_data *td)
 		sprintf(foo, "%s", tmp);
 	}
 
-	if (td->ioscheduler)
+	if (td->o.ioscheduler && !td->sysfs_root)
 		td->sysfs_root = strdup(foo);
 
 	disk_util_add(dev, foo);
 }
 
+void init_disk_util(struct thread_data *td)
+{
+	struct fio_file *f;
+	unsigned int i;
+
+	if (!td->o.do_disk_util ||
+	    (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL)))
+		return;
+
+	for_each_file(td, f, i)
+		__init_disk_util(td, f);
+}
+
 void disk_util_timer_arm(void)
 {
 	itimer.it_value.tv_sec = 0;
@@ -310,13 +330,15 @@ 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;
+
+	getrusage(RUSAGE_SELF, &ts->ru_end);
 
-	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);
+	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,
@@ -347,18 +369,18 @@ static void show_group_stats(struct group_run_stats *rs, int id)
 	const char *ddir_str[] = { "   READ", "  WRITE" };
 	int i;
 
-	fprintf(f_out, "\nRun status group %d (all jobs):\n", id);
+	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, 1);
-		p2 = num2str(rs->agg[i], 6, 1);
-		p3 = num2str(rs->min_bw[i], 6, 1);
-		p4 = num2str(rs->max_bw[i], 6, 1);
+		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);
 
-		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]);
+		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);
@@ -374,7 +396,7 @@ static void show_disk_util(void)
 	struct disk_util *du;
 	double util;
 
-	fprintf(f_out, "\nDisk stats (read/write):\n");
+	log_info("\nDisk stats (read/write):\n");
 
 	list_for_each(entry, &disk_list) {
 		du = list_entry(entry, struct disk_util, list);
@@ -384,7 +406,7 @@ static void show_disk_util(void)
 		if (util > 100.0)
 			util = 100.0;
 
-		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);
+		log_info("  %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);
 	}
 
 	/*
@@ -398,89 +420,120 @@ static void show_disk_util(void)
 	}
 }
 
-static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs,
+#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;
+	}
+}
+
+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;
+	}
+}
+
+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;
+	char *io_p, *bw_p, *iops_p;
 
-	if (!td->runtime[ddir])
+	if (!ts->runtime[ddir])
 		return;
 
-	bw = td->io_bytes[ddir] / td->runtime[ddir];
-	io_p = num2str(td->io_bytes[ddir] >> 10, 6, 1);
-	bw_p = num2str(bw, 6, 1);
+	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);
 
-	fprintf(f_out, "  %s: io=%siB, bw=%siB/s, runt=%6lumsec\n", ddir_str[ddir], io_p, bw_p, td->runtime[ddir]);
+	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(&td->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(&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->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->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(&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, stdev=%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];
-	int i;
+	double io_u_lat[FIO_IO_U_LAT_NR];
 
-	if (!(td->io_bytes[0] + td->io_bytes[1]))
+	if (!(ts->io_bytes[0] + ts->io_bytes[1]))
 		return;
 
-	if (!td->error)
-		fprintf(f_out, "%s: (groupid=%d): err=%2d: pid=%d\n",td->name, td->groupid, td->error, td->pid);
+	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
-		fprintf(f_out, "%s: (groupid=%d): err=%2d (%s): pid=%d\n",td->name, td->groupid, td->error, td->verror, td->pid);
+		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);
 
-	/*
-	 * 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;
-	}
+	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]);
+	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]);
 
-	if (td->description)
-		fprintf(f_out, "%s\n", td->description);
+	log_info("     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]);
+	log_info("     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;
@@ -488,60 +541,99 @@ 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(";%3.1f%%;%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]);
+	log_info(";%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)
+		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));
 
@@ -553,37 +645,125 @@ 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;
-
-		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;
+		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->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++) {
@@ -601,13 +781,14 @@ 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) {
@@ -618,12 +799,13 @@ 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;
@@ -631,8 +813,7 @@ 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->mean += delta / (is->samples + 1.0);
 	is->S += delta * (val - is->mean);
 
 	is->samples++;
@@ -670,38 +851,41 @@ void add_agg_sample(unsigned long val, enum fio_ddir ddir)
 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;
 
-	if (td->slat_log)
-		add_log_sample(td, td->slat_log, msec, ddir);
+	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)
 {
-	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];
 }
-
-