[fio.git] / steadystate.c

#include <stdlib.h>

#include "fio.h"
#include "steadystate.h"
#include "helper_thread.h"

void steadystate_setup()
{
	int i, prev_groupid;
	struct thread_data *td, *prev_td;

	if (!steadystate)
		return;

	/*
	 * if group reporting is enabled, identify the last td
	 * for each group and use it for storing steady state
	 * data
	 */
	prev_groupid = -1;
	prev_td = NULL;
	for_each_td(td, i) {
		if (td->ts.ss == NULL)
			continue;

		if (!td->o.group_reporting) {
			steadystate_alloc(td);
			continue;
		}

		if (prev_groupid != td->groupid) {
			if (prev_td != NULL) {
				prev_td->ss.last_in_group = 1;
				steadystate_alloc(prev_td);
			}
			prev_groupid = td->groupid;
		}
		prev_td = td;
	}

	if (prev_td != NULL && prev_td->o.group_reporting) {
		prev_td->ss.last_in_group = 1;
		steadystate_alloc(prev_td);
	}
}

void steadystate_alloc(struct thread_data *td)
{
	int i;

	td->ss.bw_data = malloc(td->ss.dur * sizeof(unsigned long));
	td->ss.iops_data = malloc(td->ss.dur * sizeof(unsigned long));
	/* initialize so that it is obvious if the cache is not full in the output */
	for (i = 0; i < td->ss.dur; i++)
		td->ss.iops_data[i] = td->ss.bw_data[i] = 0;
}

void steadystate_check()
{
	int i, j, ddir, prev_groupid, group_ramp_time_over = 0;
	unsigned long rate_time;
	struct thread_data *td, *td2;
	struct timeval now;
	unsigned long group_bw = 0, group_iops = 0;
	unsigned long long td_iops;
	unsigned long long td_bytes;

	prev_groupid = -1;
	for_each_td(td, i) {
		struct steadystate_data *ss = &td->ss;

		if (!ss->dur || td->runstate <= TD_SETTING_UP || td->runstate >= TD_EXITED || ss->attained)
			continue;

		td_iops = 0;
		td_bytes = 0;
		if (!td->o.group_reporting ||
		    (td->o.group_reporting && td->groupid != prev_groupid)) {
			group_bw = 0;
			group_iops = 0;
			group_ramp_time_over = 0;
		}
		prev_groupid = td->groupid;

		fio_gettime(&now, NULL);
		if (ss->ramp_time && !ss->ramp_time_over)
			/* 
			 * Begin recording data one second after ss->ramp_time
			 * has elapsed
			 */
			if (utime_since(&td->epoch, &now) >= (ss->ramp_time + 1000000L))
				ss->ramp_time_over = 1;

		td_io_u_lock(td);
		for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
			td_iops += td->io_blocks[ddir];
			td_bytes += td->io_bytes[ddir];
		}
		td_io_u_unlock(td);

		rate_time = mtime_since(&ss->prev_time, &now);
		memcpy(&ss->prev_time, &now, sizeof(now));

		/* 
		 * Begin monitoring when job starts but don't actually use
		 * data in checking stopping criterion until ss->ramp_time is
		 * over. This ensures that we will have a sane value in
		 * prev_iops/bw the first time through after ss->ramp_time
		 * is done.
		 */
		if (ss->ramp_time_over) {
			group_bw += 1000 * (td_bytes - ss->prev_bytes) / rate_time;
	        	group_iops += 1000 * (td_iops - ss->prev_iops) / rate_time;
			++group_ramp_time_over;
		}
		ss->prev_iops = td_iops;
		ss->prev_bytes = td_bytes;

		if (td->o.group_reporting && !ss->last_in_group)
			continue;

		/* don't begin checking criterion until ss->ramp_time is over for at least one thread in group */
		if (!group_ramp_time_over)
			continue;

		dprint(FD_STEADYSTATE, "steadystate_check() thread: %d, groupid: %u, rate_msec: %ld, iops: %lu, bw: %lu, head: %d, tail: %d\n", 
			i, td->groupid, rate_time, group_iops, group_bw, ss->head, ss->tail);

		if (ss->evaluate(group_iops, group_bw, td))
		{
			if (td->o.group_reporting)
				for_each_td(td2, j) {
					if (td2->groupid == td->groupid) {
						td2->ss.attained = 1;
						fio_mark_td_terminate(td2);
					}
				}
			else {
				ss->attained = 1;
				fio_mark_td_terminate(td);
			}
		}
	}
}

bool steadystate_slope(unsigned long iops, unsigned long bw, struct thread_data *td)
{
	int i, j;
	double result;
	struct steadystate_data *ss = &td->ss;
	unsigned long new_val = ss->check_iops ? iops : bw;

	ss->bw_data[ss->tail] = bw;
	ss->iops_data[ss->tail] = iops;

	if (ss->tail < ss->head || (ss->tail - ss->head == ss->dur - 1))
	{
		if (ss->sum_y == 0)	/* first time through */
		{
			for(i = 0; i < ss->dur; i++)
			{
				ss->sum_y += ss->check_iops ? ss->iops_data[i] : ss->bw_data[i];
				j = ss->head + i;
				if (j >= ss->dur)
					j -= ss->dur;
				ss->sum_xy += (ss->check_iops ? ss->iops_data[j] : ss->bw_data[j]) * i;
			}
		} else {		/* easy to update the sums */
			ss->sum_y -= ss->oldest_y;
			ss->sum_y += new_val;
			ss->sum_xy = ss->sum_xy - ss->sum_y + ss->dur * new_val;
		}

		ss->oldest_y = ss->check_iops ? ss->iops_data[ss->head] : ss->bw_data[ss->head];

		/*
		 * calculate slope as (sum_xy - sum_x * sum_y / n) / (sum_(x^2) - (sum_x)^2 / n)
		 * This code assumes that all x values are equally spaced when they are often
		 * off by a few milliseconds. This assumption greatly simplifies the
		 * calculations.
		 */
		ss->slope = (ss->sum_xy - (double) ss->sum_x * ss->sum_y / ss->dur) / (ss->sum_x_sq - (double) ss->sum_x * ss->sum_x / ss->dur);
		ss->criterion = ss->pct ? 100.0 * ss->slope / (ss->sum_y / ss->dur) : ss->slope;

		dprint(FD_STEADYSTATE, "sum_y: %llu, sum_xy: %llu, slope: %f, criterion: %f, limit: %f\n",
			ss->sum_y, ss->sum_xy, ss->slope, ss->criterion, ss->limit);

		result = ss->criterion * (ss->criterion < 0.0 ? -1.0 : 1.0);
		if (result < ss->limit)
			return true;
	}

	ss->tail = (ss->tail + 1) % ss->dur;
	if (ss->tail <= ss->head)
		ss->head = (ss->head + 1) % ss->dur;
	return false;
}

bool steadystate_deviation(unsigned long iops, unsigned long bw, struct thread_data *td)
{
	int i;
	double diff;
	double mean;

	struct steadystate_data *ss = &td->ss;

	ss->bw_data[ss->tail] = bw;
	ss->iops_data[ss->tail] = iops;

	if (ss->tail < ss->head || (ss->tail - ss->head == ss->dur - 1))
	{
		if (ss->sum_y == 0)	/* first time through */
		{
			for(i = 0; i < ss->dur; i++)
				ss->sum_y += ss->check_iops ? ss->iops_data[i] : ss->bw_data[i];
		} else {		/* easy to update the sum */
			ss->sum_y -= ss->oldest_y;
			ss->sum_y += ss->check_iops ? ss->iops_data[ss->tail] : ss->bw_data[ss->tail];
		}

		ss->oldest_y = ss->check_iops ? ss->iops_data[ss->head] : ss->bw_data[ss->head];
		mean = (double) ss->sum_y / ss->dur;
		ss->deviation = 0.0;

		for (i = 0; i < ss->dur; i++)
		{	
			diff = (double) (ss->check_iops ? ss->iops_data[i] : ss->bw_data[i]) - mean;
			ss->deviation = max(ss->deviation, diff * (diff < 0.0 ? -1.0 : 1.0));
		}

		ss->criterion = ss->pct ? 100.0 * ss->deviation / mean : ss->deviation;

		dprint(FD_STEADYSTATE, "sum_y: %llu, mean: %f, max diff: %f, objective: %f, limit: %f\n", ss->sum_y, mean, ss->deviation, ss->criterion, ss->limit);

		if (ss->criterion < ss->limit)
			return true;
	}

	ss->tail = (ss->tail + 1) % ss->dur;
	if (ss->tail <= ss->head)
		ss->head = (ss->head + 1) % ss->dur;
	return false;
}
Commit	Line	Data
ba8fb6f6 VF	1	#include <stdlib.h>
ba8fb6f6 VF	2
16e56d25 VF	3	#include "fio.h"
	4	#include "steadystate.h"
	5	#include "helper_thread.h"
	6
	7	void steadystate_setup()
	8	{
	9	int i, prev_groupid;
	10	struct thread_data td, prev_td;
	11
	12	if (!steadystate)
	13	return;
	14
	15	/*
	16	* if group reporting is enabled, identify the last td
	17	* for each group and use it for storing steady state
	18	* data
	19	*/
	20	prev_groupid = -1;
	21	prev_td = NULL;
	22	for_each_td(td, i) {
ba8fb6f6 VF	23	if (td->ts.ss == NULL)
	24	continue;
	25
	26	if (!td->o.group_reporting) {
	27	steadystate_alloc(td);
16e56d25	28	continue;
ba8fb6f6	29	}
16e56d25 VF	30
16e56d25 VF	31	if (prev_groupid != td->groupid) {
ba8fb6f6	32	if (prev_td != NULL) {
16e56d25	33	prev_td->ss.last_in_group = 1;
ba8fb6f6 VF	34	steadystate_alloc(prev_td);
ba8fb6f6 VF	35	}
16e56d25 VF	36	prev_groupid = td->groupid;
	37	}
	38	prev_td = td;
	39	}
	40
ba8fb6f6	41	if (prev_td != NULL && prev_td->o.group_reporting) {
16e56d25	42	prev_td->ss.last_in_group = 1;
ba8fb6f6 VF	43	steadystate_alloc(prev_td);
	44	}
	45	}
	46
	47	void steadystate_alloc(struct thread_data *td)
	48	{
	49	int i;
	50
	51	td->ss.bw_data = malloc(td->ss.dur * sizeof(unsigned long));
	52	td->ss.iops_data = malloc(td->ss.dur * sizeof(unsigned long));
	53	/* initialize so that it is obvious if the cache is not full in the output */
	54	for (i = 0; i < td->ss.dur; i++)
	55	td->ss.iops_data[i] = td->ss.bw_data[i] = 0;
16e56d25 VF	56	}
	57
	58	void steadystate_check()
	59	{
	60	int i, j, ddir, prev_groupid, group_ramp_time_over = 0;
	61	unsigned long rate_time;
	62	struct thread_data td, td2;
	63	struct timeval now;
	64	unsigned long group_bw = 0, group_iops = 0;
	65	unsigned long long td_iops;
	66	unsigned long long td_bytes;
	67
	68	prev_groupid = -1;
	69	for_each_td(td, i) {
	70	struct steadystate_data *ss = &td->ss;
	71
	72	if (!ss->dur \|\| td->runstate <= TD_SETTING_UP \|\| td->runstate >= TD_EXITED \|\| ss->attained)
	73	continue;
	74
	75	td_iops = 0;
	76	td_bytes = 0;
	77	if (!td->o.group_reporting \|\|
	78	(td->o.group_reporting && td->groupid != prev_groupid)) {
	79	group_bw = 0;
	80	group_iops = 0;
	81	group_ramp_time_over = 0;
	82	}
	83	prev_groupid = td->groupid;
	84
	85	fio_gettime(&now, NULL);
	86	if (ss->ramp_time && !ss->ramp_time_over)
	87	/*
	88	* Begin recording data one second after ss->ramp_time
	89	* has elapsed
	90	*/
	91	if (utime_since(&td->epoch, &now) >= (ss->ramp_time + 1000000L))
	92	ss->ramp_time_over = 1;
	93
0587ef32	94	td_io_u_lock(td);
16e56d25 VF	95	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
	96	td_iops += td->io_blocks[ddir];
	97	td_bytes += td->io_bytes[ddir];
	98	}
0587ef32	99	td_io_u_unlock(td);
16e56d25 VF	100
	101	rate_time = mtime_since(&ss->prev_time, &now);
	102	memcpy(&ss->prev_time, &now, sizeof(now));
	103
	104	/*
	105	* Begin monitoring when job starts but don't actually use
	106	* data in checking stopping criterion until ss->ramp_time is
	107	* over. This ensures that we will have a sane value in
	108	* prev_iops/bw the first time through after ss->ramp_time
	109	* is done.
	110	*/
	111	if (ss->ramp_time_over) {
	112	group_bw += 1000 * (td_bytes - ss->prev_bytes) / rate_time;
	113	group_iops += 1000 * (td_iops - ss->prev_iops) / rate_time;
	114	++group_ramp_time_over;
	115	}
	116	ss->prev_iops = td_iops;
	117	ss->prev_bytes = td_bytes;
	118
	119	if (td->o.group_reporting && !ss->last_in_group)
	120	continue;
	121
	122	/* don't begin checking criterion until ss->ramp_time is over for at least one thread in group */
	123	if (!group_ramp_time_over)
	124	continue;
	125
	126	dprint(FD_STEADYSTATE, "steadystate_check() thread: %d, groupid: %u, rate_msec: %ld, iops: %lu, bw: %lu, head: %d, tail: %d\n",
	127	i, td->groupid, rate_time, group_iops, group_bw, ss->head, ss->tail);
	128
	129	if (ss->evaluate(group_iops, group_bw, td))
	130	{
	131	if (td->o.group_reporting)
	132	for_each_td(td2, j) {
	133	if (td2->groupid == td->groupid) {
	134	td2->ss.attained = 1;
	135	fio_mark_td_terminate(td2);
	136	}
	137	}
	138	else {
	139	ss->attained = 1;
	140	fio_mark_td_terminate(td);
	141	}
	142	}
	143	}
	144	}
	145
	146	bool steadystate_slope(unsigned long iops, unsigned long bw, struct thread_data *td)
	147	{
ba8fb6f6	148	int i, j;
16e56d25	149	double result;
16e56d25	150	struct steadystate_data *ss = &td->ss;
ba8fb6f6	151	unsigned long new_val = ss->check_iops ? iops : bw;
16e56d25	152
ba8fb6f6 VF	153	ss->bw_data[ss->tail] = bw;
ba8fb6f6 VF	154	ss->iops_data[ss->tail] = iops;
16e56d25 VF	155
	156	if (ss->tail < ss->head \|\| (ss->tail - ss->head == ss->dur - 1))
	157	{
	158	if (ss->sum_y == 0) /* first time through */
	159	{
	160	for(i = 0; i < ss->dur; i++)
	161	{
ba8fb6f6 VF	162	ss->sum_y += ss->check_iops ? ss->iops_data[i] : ss->bw_data[i];
	163	j = ss->head + i;
	164	if (j >= ss->dur)
	165	j -= ss->dur;
	166	ss->sum_xy += (ss->check_iops ? ss->iops_data[j] : ss->bw_data[j]) * i;
16e56d25 VF	167	}
	168	} else { /* easy to update the sums */
	169	ss->sum_y -= ss->oldest_y;
ba8fb6f6 VF	170	ss->sum_y += new_val;
ba8fb6f6 VF	171	ss->sum_xy = ss->sum_xy - ss->sum_y + ss->dur * new_val;
16e56d25 VF	172	}
16e56d25 VF	173
ba8fb6f6	174	ss->oldest_y = ss->check_iops ? ss->iops_data[ss->head] : ss->bw_data[ss->head];
16e56d25 VF	175
	176	/*
	177	* calculate slope as (sum_xy - sum_x * sum_y / n) / (sum_(x^2) - (sum_x)^2 / n)
	178	* This code assumes that all x values are equally spaced when they are often
	179	* off by a few milliseconds. This assumption greatly simplifies the
	180	* calculations.
	181	*/
ba8fb6f6	182	ss->slope = (ss->sum_xy - (double) ss->sum_x * ss->sum_y / ss->dur) / (ss->sum_x_sq - (double) ss->sum_x * ss->sum_x / ss->dur);
6da94b07	183	ss->criterion = ss->pct ? 100.0 * ss->slope / (ss->sum_y / ss->dur) : ss->slope;
16e56d25 VF	184
16e56d25 VF	185	dprint(FD_STEADYSTATE, "sum_y: %llu, sum_xy: %llu, slope: %f, criterion: %f, limit: %f\n",
ba8fb6f6	186	ss->sum_y, ss->sum_xy, ss->slope, ss->criterion, ss->limit);
16e56d25	187
6da94b07	188	result = ss->criterion * (ss->criterion < 0.0 ? -1.0 : 1.0);
16e56d25 VF	189	if (result < ss->limit)
	190	return true;
	191	}
	192
	193	ss->tail = (ss->tail + 1) % ss->dur;
	194	if (ss->tail <= ss->head)
	195	ss->head = (ss->head + 1) % ss->dur;
	196	return false;
	197	}
	198
	199	bool steadystate_deviation(unsigned long iops, unsigned long bw, struct thread_data *td)
	200	{
	201	int i;
	202	double diff;
	203	double mean;
16e56d25 VF	204
	205	struct steadystate_data *ss = &td->ss;
	206
ba8fb6f6 VF	207	ss->bw_data[ss->tail] = bw;
ba8fb6f6 VF	208	ss->iops_data[ss->tail] = iops;
16e56d25 VF	209
	210	if (ss->tail < ss->head \|\| (ss->tail - ss->head == ss->dur - 1))
	211	{
	212	if (ss->sum_y == 0) /* first time through */
	213	{
	214	for(i = 0; i < ss->dur; i++)
ba8fb6f6	215	ss->sum_y += ss->check_iops ? ss->iops_data[i] : ss->bw_data[i];
16e56d25 VF	216	} else { /* easy to update the sum */
16e56d25 VF	217	ss->sum_y -= ss->oldest_y;
ba8fb6f6	218	ss->sum_y += ss->check_iops ? ss->iops_data[ss->tail] : ss->bw_data[ss->tail];
16e56d25 VF	219	}
16e56d25 VF	220
ba8fb6f6	221	ss->oldest_y = ss->check_iops ? ss->iops_data[ss->head] : ss->bw_data[ss->head];
16e56d25	222	mean = (double) ss->sum_y / ss->dur;
ba8fb6f6	223	ss->deviation = 0.0;
16e56d25 VF	224
	225	for (i = 0; i < ss->dur; i++)
	226	{
ba8fb6f6	227	diff = (double) (ss->check_iops ? ss->iops_data[i] : ss->bw_data[i]) - mean;
6da94b07	228	ss->deviation = max(ss->deviation, diff * (diff < 0.0 ? -1.0 : 1.0));
16e56d25 VF	229	}
16e56d25 VF	230
6da94b07	231	ss->criterion = ss->pct ? 100.0 * ss->deviation / mean : ss->deviation;
16e56d25	232
ba8fb6f6	233	dprint(FD_STEADYSTATE, "sum_y: %llu, mean: %f, max diff: %f, objective: %f, limit: %f\n", ss->sum_y, mean, ss->deviation, ss->criterion, ss->limit);
16e56d25 VF	234
	235	if (ss->criterion < ss->limit)
	236	return true;
	237	}
	238
	239	ss->tail = (ss->tail + 1) % ss->dur;
	240	if (ss->tail <= ss->head)
	241	ss->head = (ss->head + 1) % ss->dur;
	242	return false;
	243	}