Drop workload phase recording and slow telemetry sampling to 2/sec
The telemetry thread's per-core /sys scan was contending with the timed benchmark passes, and could corrupt the workload-size calibration step, causing run-to-run swings of up to 4x. Removing phase recording and dropping the sample rate from 4/sec to 2/sec brings runs back to <5% variance.
This commit is contained in:
+4
-27
@@ -173,6 +173,8 @@ extern uint64_t fb_c_chase(void **ptrs, uint64_t steps);
|
||||
# define REPEATS 3 /* Number of tries. */
|
||||
#endif
|
||||
|
||||
#define TELEMETRY_REFRESH_INTERVAL 0.5
|
||||
|
||||
/* Simple repeatable random numbers. */
|
||||
|
||||
static uint64_t rng_state = 0x853c49e6748fea9bULL;
|
||||
@@ -220,25 +222,17 @@ struct background_metrics {
|
||||
int samples, available;
|
||||
};
|
||||
|
||||
|
||||
#define MAX_TELEMETRY_SAMPLES 1800
|
||||
#define MAX_TELEMETRY_PHASES 44
|
||||
|
||||
struct telemetry_sample {
|
||||
uint64_t elapsed_ms;
|
||||
double temperature_c, clock_mhz;
|
||||
};
|
||||
|
||||
struct telemetry_phase {
|
||||
uint64_t elapsed_ms;
|
||||
const char *suite, *mode;
|
||||
int workload_index;
|
||||
};
|
||||
|
||||
struct telemetry_monitor {
|
||||
struct telemetry_sample samples[MAX_TELEMETRY_SAMPLES];
|
||||
size_t count;
|
||||
struct telemetry_phase phases[MAX_TELEMETRY_PHASES];
|
||||
size_t phase_count;
|
||||
double started;
|
||||
volatile int stop;
|
||||
#if defined(_WIN32)
|
||||
@@ -248,19 +242,6 @@ struct telemetry_monitor {
|
||||
#endif
|
||||
};
|
||||
|
||||
static void record_telemetry_phase(struct telemetry_monitor *monitor,
|
||||
const char *suite, int workload_index,
|
||||
const char *mode)
|
||||
{
|
||||
struct telemetry_phase *phase;
|
||||
if (monitor->phase_count >= MAX_TELEMETRY_PHASES) return;
|
||||
phase = &monitor->phases[monitor->phase_count++];
|
||||
phase->elapsed_ms = (uint64_t)((now_seconds() - monitor->started) * 1000.0);
|
||||
phase->suite = suite;
|
||||
phase->workload_index = workload_index;
|
||||
phase->mode = mode;
|
||||
}
|
||||
|
||||
static int read_number_file(const char *path, double *value)
|
||||
{
|
||||
FILE *f = fopen(path, "r");
|
||||
@@ -420,7 +401,7 @@ static void *telemetry_entry(void *arg)
|
||||
double next = monitor->started;
|
||||
while (!monitor->stop) {
|
||||
double current = now_seconds();
|
||||
if (current >= next) { record_telemetry_sample(monitor); next += 0.25; }
|
||||
if (current >= next) { record_telemetry_sample(monitor); next += TELEMETRY_REFRESH_INTERVAL; }
|
||||
telemetry_sleep();
|
||||
}
|
||||
record_telemetry_sample(monitor);
|
||||
@@ -1063,9 +1044,7 @@ int fossbench_run(int verbose, int upload_mode, int system_check)
|
||||
fflush(stdout);
|
||||
|
||||
/* Run every test with all cores and one core. */
|
||||
record_telemetry_phase(&telemetry, "asm", (int)i, "multicore");
|
||||
raw_multi[i] = run_test(&tests[i], (int)g_ncores);
|
||||
record_telemetry_phase(&telemetry, "asm", (int)i, "singlecore");
|
||||
raw_single[i] = run_test(&tests[i], 1);
|
||||
|
||||
printf(" %12.1f %-11s %7.2fs\n",
|
||||
@@ -1086,9 +1065,7 @@ int fossbench_run(int verbose, int upload_mode, int system_check)
|
||||
for (i = 0; i < NTESTS; i++) {
|
||||
printf(" %-24s", tests[i].name);
|
||||
fflush(stdout);
|
||||
record_telemetry_phase(&telemetry, "c", (int)i, "multicore");
|
||||
real_multi[i] = run_test(&tests[i], (int)g_ncores);
|
||||
record_telemetry_phase(&telemetry, "c", (int)i, "singlecore");
|
||||
real_single[i] = run_test(&tests[i], 1);
|
||||
printf(" %12.1f %-11s %7.2fs\n",
|
||||
display_metric(&tests[i], &real_multi[i]), tests[i].unit,
|
||||
|
||||
@@ -246,21 +246,6 @@ static int upload_results(const struct system_info *info,
|
||||
used += (size_t)n;
|
||||
}
|
||||
}
|
||||
n = snprintf(payload + used, sizeof payload - used, "],\"telemetry_phases\":[");
|
||||
if (n < 0 || (size_t)n >= sizeof payload - used) return 0;
|
||||
used += (size_t)n;
|
||||
{
|
||||
size_t i;
|
||||
for (i = 0; i < telemetry->phase_count; i++) {
|
||||
const struct telemetry_phase *phase = &telemetry->phases[i];
|
||||
n = snprintf(payload + used, sizeof payload - used,
|
||||
"%s{\"elapsed_ms\":%llu,\"suite\":\"%s\",\"workload_index\":%d,\"mode\":\"%s\"}",
|
||||
i ? "," : "", (unsigned long long)phase->elapsed_ms,
|
||||
phase->suite, phase->workload_index, phase->mode);
|
||||
if (n < 0 || (size_t)n >= sizeof payload - used) return 0;
|
||||
used += (size_t)n;
|
||||
}
|
||||
}
|
||||
if (used + 4 >= sizeof payload) return 0;
|
||||
memcpy(payload + used, "]}}", 4);
|
||||
payload_len = (int)(used + 3);
|
||||
|
||||
Reference in New Issue
Block a user