+#include <stdlib.h>
+
#include "fio.h"
#include "steadystate.h"
#include "helper_thread.h"
-void steadystate_setup()
+bool steadystate_enabled = false;
+
+static void steadystate_alloc(struct thread_data *td)
+{
+ int i;
+
+ td->ss.bw_data = malloc(td->ss.dur * sizeof(uint64_t));
+ td->ss.iops_data = malloc(td->ss.dur * sizeof(uint64_t));
+ /* 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;
+
+ td->ss.state |= __FIO_SS_DATA;
+}
+
+void steadystate_setup(void)
{
int i, prev_groupid;
struct thread_data *td, *prev_td;
- if (!steadystate)
+ if (!steadystate_enabled)
return;
/*
prev_groupid = -1;
prev_td = NULL;
for_each_td(td, i) {
- if (!td->o.group_reporting)
+ if (!td->ss.dur)
+ 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;
+ if (prev_td != NULL) {
+ 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;
+ if (prev_td != NULL && prev_td->o.group_reporting) {
+ steadystate_alloc(prev_td);
+ }
+}
+
+static bool steadystate_slope(uint64_t iops, uint64_t bw,
+ struct thread_data *td)
+{
+ int i, j;
+ double result;
+ struct steadystate_data *ss = &td->ss;
+ uint64_t new_val;
+
+ ss->bw_data[ss->tail] = bw;
+ ss->iops_data[ss->tail] = iops;
+
+ if (ss->state & __FIO_SS_IOPS)
+ new_val = iops;
+ else
+ new_val = bw;
+
+ if (ss->state & __FIO_SS_BUFFER_FULL || ss->tail - ss->head == ss->dur - 1) {
+ if (!(ss->state & __FIO_SS_BUFFER_FULL)) {
+ /* first time through */
+ for(i = 0, ss->sum_y = 0; i < ss->dur; i++) {
+ if (ss->state & __FIO_SS_IOPS)
+ ss->sum_y += ss->iops_data[i];
+ else
+ ss->sum_y += ss->bw_data[i];
+ j = (ss->head + i) % ss->dur;
+ if (ss->state & __FIO_SS_IOPS)
+ ss->sum_xy += i * ss->iops_data[j];
+ else
+ ss->sum_xy += i * ss->bw_data[j];
+ }
+ ss->state |= __FIO_SS_BUFFER_FULL;
+ } 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;
+ }
+
+ if (ss->state & __FIO_SS_IOPS)
+ ss->oldest_y = ss->iops_data[ss->head];
+ else
+ ss->oldest_y = 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);
+ if (ss->state & __FIO_SS_PCT)
+ ss->criterion = 100.0 * ss->slope / (ss->sum_y / ss->dur);
+ else
+ ss->criterion = ss->slope;
+
+ dprint(FD_STEADYSTATE, "sum_y: %llu, sum_xy: %llu, slope: %f, "
+ "criterion: %f, limit: %f\n",
+ (unsigned long long) ss->sum_y,
+ (unsigned long long) 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;
}
-void steadystate_check()
+static bool steadystate_deviation(uint64_t iops, uint64_t 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->state & __FIO_SS_BUFFER_FULL || ss->tail - ss->head == ss->dur - 1) {
+ if (!(ss->state & __FIO_SS_BUFFER_FULL)) {
+ /* first time through */
+ for(i = 0, ss->sum_y = 0; i < ss->dur; i++)
+ if (ss->state & __FIO_SS_IOPS)
+ ss->sum_y += ss->iops_data[i];
+ else
+ ss->sum_y += ss->bw_data[i];
+ ss->state |= __FIO_SS_BUFFER_FULL;
+ } else { /* easy to update the sum */
+ ss->sum_y -= ss->oldest_y;
+ if (ss->state & __FIO_SS_IOPS)
+ ss->sum_y += ss->iops_data[ss->tail];
+ else
+ ss->sum_y += ss->bw_data[ss->tail];
+ }
+
+ if (ss->state & __FIO_SS_IOPS)
+ ss->oldest_y = ss->iops_data[ss->head];
+ else
+ ss->oldest_y = ss->bw_data[ss->head];
+
+ mean = (double) ss->sum_y / ss->dur;
+ ss->deviation = 0.0;
+
+ for (i = 0; i < ss->dur; i++) {
+ if (ss->state & __FIO_SS_IOPS)
+ diff = ss->iops_data[i] - mean;
+ else
+ diff = ss->bw_data[i] - mean;
+ ss->deviation = max(ss->deviation, diff * (diff < 0.0 ? -1.0 : 1.0));
+ }
+
+ if (ss->state & __FIO_SS_PCT)
+ ss->criterion = 100.0 * ss->deviation / mean;
+ else
+ ss->criterion = ss->deviation;
+
+ dprint(FD_STEADYSTATE, "sum_y: %llu, mean: %f, max diff: %f, "
+ "objective: %f, limit: %f\n",
+ (unsigned long long) 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;
+}
+
+void steadystate_check(void)
{
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;
+ uint64_t group_bw = 0, group_iops = 0;
+ uint64_t td_iops, td_bytes;
+ bool ret;
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)
+ if (!ss->dur || td->runstate <= TD_SETTING_UP ||
+ td->runstate >= TD_EXITED || !ss->state ||
+ ss->state & __FIO_SS_ATTAINED)
continue;
td_iops = 0;
prev_groupid = td->groupid;
fio_gettime(&now, NULL);
- if (ss->ramp_time && !ss->ramp_time_over)
- /*
+ if (ss->ramp_time && !(ss->state & __FIO_SS_RAMP_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;
+ ss->state |= __FIO_SS_RAMP_OVER;
+ }
td_io_u_lock(td);
for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
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) {
+ if (ss->state & __FIO_SS_RAMP_OVER) {
group_bw += 1000 * (td_bytes - ss->prev_bytes) / rate_time;
- group_iops += 1000 * (td_iops - ss->prev_iops) / 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)
+ if (td->o.group_reporting && !(ss->state & __FIO_SS_DATA))
continue;
- /* don't begin checking criterion until ss->ramp_time is over for at least one thread in group */
+ /*
+ * 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)
+ dprint(FD_STEADYSTATE, "steadystate_check() thread: %d, "
+ "groupid: %u, rate_msec: %ld, "
+ "iops: %llu, bw: %llu, head: %d, tail: %d\n",
+ i, td->groupid, rate_time,
+ (unsigned long long) group_iops,
+ (unsigned long long) group_bw,
+ ss->head, ss->tail);
+
+ if (ss->state & __FIO_SS_SLOPE)
+ ret = steadystate_slope(group_iops, group_bw, td);
+ else
+ ret = steadystate_deviation(group_iops, group_bw, td);
+
+ if (ret) {
+ if (td->o.group_reporting) {
for_each_td(td2, j) {
if (td2->groupid == td->groupid) {
- td2->ss.attained = 1;
+ td2->ss.state |= __FIO_SS_ATTAINED;
fio_mark_td_terminate(td2);
}
}
- else {
- ss->attained = 1;
+ } else {
+ ss->state |= __FIO_SS_ATTAINED;
fio_mark_td_terminate(td);
}
}
}
}
-bool steadystate_slope(unsigned long iops, unsigned long bw, struct thread_data *td)
+int td_steadystate_init(struct thread_data *td)
{
- int i, x;
- double result;
- double slope;
struct steadystate_data *ss = &td->ss;
+ struct thread_options *o = &td->o;
+ struct thread_data *td2;
+ int j;
- ss->cache[ss->tail] = ss->check_iops ? iops : bw;
-
- 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->cache[i];
- x = ss->head + i;
- if (x >= ss->dur)
- x -= ss->dur;
- ss->sum_xy += ss->cache[x] * i;
- }
- } else { /* easy to update the sums */
- ss->sum_y -= ss->oldest_y;
- ss->sum_y += ss->cache[ss->tail];
- ss->sum_xy = ss->sum_xy - ss->sum_y + ss->dur * ss->cache[ss->tail];
- }
+ memset(ss, 0, sizeof(*ss));
- ss->oldest_y = ss->cache[ss->head];
+ if (o->ss_dur) {
+ steadystate_enabled = true;
+ o->ss_dur /= 1000000L;
- /*
- * 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.
- */
- 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 ? slope / (ss->sum_y / ss->dur) * 100.0: slope;
+ /* put all steady state info in one place */
+ ss->dur = o->ss_dur;
+ ss->limit = o->ss_limit.u.f;
+ ss->ramp_time = o->ss_ramp_time;
- dprint(FD_STEADYSTATE, "sum_y: %llu, sum_xy: %llu, slope: %f, criterion: %f, limit: %f\n",
- ss->sum_y, ss->sum_xy, slope, ss->criterion, ss->limit);
+ ss->state = o->ss_state;
+ if (!td->ss.ramp_time)
+ ss->state |= __FIO_SS_RAMP_OVER;
- result = ss->criterion * (ss->criterion < 0.0 ? -1 : 1);
- if (result < ss->limit)
- return true;
+ ss->sum_x = o->ss_dur * (o->ss_dur - 1) / 2;
+ ss->sum_x_sq = (o->ss_dur - 1) * (o->ss_dur) * (2*o->ss_dur - 1) / 6;
}
- ss->tail = (ss->tail + 1) % ss->dur;
- if (ss->tail <= ss->head)
- ss->head = (ss->head + 1) % ss->dur;
- return false;
+ /* make sure that ss options are consistent within reporting group */
+ for_each_td(td2, j) {
+ if (td2->groupid == td->groupid) {
+ struct steadystate_data *ss2 = &td2->ss;
+
+ if (ss2->dur != ss->dur ||
+ ss2->limit != ss->limit ||
+ ss2->ramp_time != ss->ramp_time ||
+ ss2->state != ss->state ||
+ ss2->sum_x != ss->sum_x ||
+ ss2->sum_x_sq != ss->sum_x_sq) {
+ td_verror(td, EINVAL, "job rejected: steadystate options must be consistent within reporting groups");
+ return 1;
+ }
+ }
+ }
+
+ return 0;
}
-bool steadystate_deviation(unsigned long iops, unsigned long bw, struct thread_data *td)
+uint64_t steadystate_bw_mean(struct thread_stat *ts)
{
int i;
- double diff;
- double mean;
- double deviation;
-
- struct steadystate_data *ss = &td->ss;
-
- ss->cache[ss->tail] = ss->check_iops ? iops : bw;
-
- 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->cache[i];
- } else { /* easy to update the sum */
- ss->sum_y -= ss->oldest_y;
- ss->sum_y += ss->cache[ss->tail];
- }
+ uint64_t sum;
- ss->oldest_y = ss->cache[ss->head];
- mean = (double) ss->sum_y / ss->dur;
- deviation = 0.0;
+ for (i = 0, sum = 0; i < ts->ss_dur; i++)
+ sum += ts->ss_bw_data[i];
- for (i = 0; i < ss->dur; i++)
- {
- diff = (double) ss->cache[i] - mean;
- deviation = max(deviation, diff * (diff < 0.0 ? -1 : 1));
- }
+ return sum / ts->ss_dur;
+}
- ss->criterion = ss->pct ? deviation / mean * 100.0 : deviation;
+uint64_t steadystate_iops_mean(struct thread_stat *ts)
+{
+ int i;
+ uint64_t sum;
- dprint(FD_STEADYSTATE, "sum_y: %llu, mean: %f, max diff: %f, objective: %f, limit: %f\n", ss->sum_y, mean, deviation, ss->criterion, ss->limit);
+ for (i = 0, sum = 0; i < ts->ss_dur; i++)
+ sum += ts->ss_iops_data[i];
- 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;
+ return sum / ts->ss_dur;
}