Files
chrony/quantiles.c
Miroslav Lichvar 2da4e3ce53 quantiles: force step update with stable input values
The algorithm was designed for estimating quantiles in streams of
integer values. When the estimate is equal to the input value, the
step state variable does not change. This causes problems for the
floating-point adaptation used for measurents of delay in chrony.

One problem is numerical instability due to the strict comparison of
the input value and the current estimate.

Another problem is with signals that are so stable that the nanosecond
resolution of the system functions becomes the limitation. There is a
large difference in the value of the step state variable, which
determines how quickly the estimate will adapt to a new distribution,
between signals that are constant in the nanosecond resolution and
signals that can move in two nanoseconds.

Change the estimate update to never consider the input value equal to
the current estimate and don't set the estimate exactly to the input
value. Keep it off by a quarter of the minimum step to force jumping
around the input value if it's constant and decreasing the step variable
to negative values. Also fix the initial adjustment to step at least by
the minimum step (the original algorithm is described with ceil(), not
fabs()).
2024-11-21 15:59:56 +01:00

230 lines
5.7 KiB
C

/*
chronyd/chronyc - Programs for keeping computer clocks accurate.
**********************************************************************
* Copyright (C) Miroslav Lichvar 2022
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of version 2 of the GNU General Public License as
* published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*
**********************************************************************
=======================================================================
Estimation of quantiles using the Frugal-2U streaming algorithm
(https://arxiv.org/pdf/1407.1121v1.pdf)
*/
#include "config.h"
#include "logging.h"
#include "memory.h"
#include "quantiles.h"
#include "regress.h"
#include "util.h"
/* Maximum number of repeated estimates for stabilisation */
#define MAX_REPEAT 64
struct Quantile {
double est;
double step;
int sign;
};
struct QNT_Instance_Record {
struct Quantile *quants;
int n_quants;
int repeat;
int q;
int min_k;
double min_step;
int n_set;
};
/* ================================================== */
QNT_Instance
QNT_CreateInstance(int min_k, int max_k, int q, int repeat, double min_step)
{
QNT_Instance inst;
long seed;
if (q < 2 || min_k > max_k || min_k < 1 || max_k >= q ||
repeat < 1 || repeat > MAX_REPEAT || min_step <= 0.0)
assert(0);
inst = MallocNew(struct QNT_Instance_Record);
inst->n_quants = (max_k - min_k + 1) * repeat;
inst->quants = MallocArray(struct Quantile, inst->n_quants);
inst->repeat = repeat;
inst->q = q;
inst->min_k = min_k;
inst->min_step = min_step;
QNT_Reset(inst);
/* Seed the random number generator, which will not be isolated from
other instances and other random() users */
UTI_GetRandomBytes(&seed, sizeof (seed));
srandom(seed);
return inst;
}
/* ================================================== */
void
QNT_DestroyInstance(QNT_Instance inst)
{
Free(inst->quants);
Free(inst);
}
/* ================================================== */
void
QNT_Reset(QNT_Instance inst)
{
int i;
inst->n_set = 0;
for (i = 0; i < inst->n_quants; i++) {
inst->quants[i].est = 0.0;
inst->quants[i].step = inst->min_step;
inst->quants[i].sign = 1;
}
}
/* ================================================== */
static void
insert_initial_value(QNT_Instance inst, double value)
{
int i, j, r = inst->repeat;
if (inst->n_set * r >= inst->n_quants)
assert(0);
/* Keep the initial estimates repeated and ordered */
for (i = inst->n_set; i > 0 && inst->quants[(i - 1) * r].est > value; i--) {
for (j = 0; j < r; j++)
inst->quants[i * r + j].est = inst->quants[(i - 1) * r].est;
}
for (j = 0; j < r; j++)
inst->quants[i * r + j].est = value;
inst->n_set++;
/* Duplicate the largest value in unset quantiles */
for (i = inst->n_set * r; i < inst->n_quants; i++)
inst->quants[i].est = inst->quants[i - 1].est;
}
/* ================================================== */
static void
update_estimate(struct Quantile *quantile, double value, double p, double rand,
double min_step)
{
if (value >= quantile->est) {
if (rand < (1.0 - p))
return;
quantile->step += quantile->sign > 0 ? min_step : -min_step;
quantile->est += quantile->step > min_step ? quantile->step : min_step;
if (quantile->est > value) {
quantile->step += value - quantile->est;
quantile->est = value + min_step / 4.0;
}
if (quantile->sign < 0 && quantile->step > min_step)
quantile->step = min_step;
quantile->sign = 1;
} else {
if (rand < p)
return;
quantile->step += quantile->sign < 0 ? min_step : -min_step;
quantile->est -= quantile->step > min_step ? quantile->step : min_step;
if (quantile->est < value) {
quantile->step += quantile->est - value;
quantile->est = value - min_step / 4.0;
}
if (quantile->sign > 0 && quantile->step > min_step)
quantile->step = min_step;
quantile->sign = -1;
}
}
/* ================================================== */
void
QNT_Accumulate(QNT_Instance inst, double value)
{
double p, rand;
int i;
/* Initialise the estimates with first received values */
if (inst->n_set * inst->repeat < inst->n_quants) {
insert_initial_value(inst, value);
return;
}
for (i = 0; i < inst->n_quants; i++) {
p = (double)(i / inst->repeat + inst->min_k) / inst->q;
rand = (double)random() / ((1U << 31) - 1);
update_estimate(&inst->quants[i], value, p, rand, inst->min_step);
}
}
/* ================================================== */
int
QNT_GetMinK(QNT_Instance inst)
{
return inst->min_k;
}
/* ================================================== */
int
QNT_GetMaxK(QNT_Instance inst)
{
return inst->min_k + (inst->n_quants / inst->repeat) - 1;
}
/* ================================================== */
double
QNT_GetMinStep(QNT_Instance inst)
{
return inst->min_step;
}
/* ================================================== */
double
QNT_GetQuantile(QNT_Instance inst, int k)
{
double estimates[MAX_REPEAT];
int i;
if (k < inst->min_k || (k - inst->min_k) * inst->repeat >= inst->n_quants)
assert(0);
for (i = 0; i < inst->repeat; i++)
estimates[i] = inst->quants[(k - inst->min_k) * inst->repeat + i].est;
return RGR_FindMedian(estimates, inst->repeat);
}