The event-spaced observation series MSER-5 and the CI both read. More...

#include <line/solvers/ldes/ldes_stats.h>

Collaboration diagram for line::ldes::engine::Observations:

Public Member Functions
	Observations (std::size_t M, std::size_t K, bool mser_on, int batch)
void	collect (const Accum &acc, double now)
	Record one observation.
Truncation	truncate () const
	The truncation point, on the AGGREGATE queue length first.

Public Attributes
bool	enabled = true
int	batch_size = 5
std::vector< double >	time
std::vector< std::vector< std::vector< double > > >	qlen
std::vector< std::vector< std::vector< double > > >	qt
std::vector< std::vector< std::vector< double > > >	bt
std::vector< std::vector< std::vector< double > > >	cmp
std::vector< std::vector< std::vector< double > > >	drp
std::vector< std::vector< double > >	last_qt
std::vector< char >	in_mser
	Whether station i feeds the truncation criterion; every one is still recorded.
double	last_time = 0.0
std::size_t	nstations = 0
std::size_t	nclasses = 0

Detailed Description

The event-spaced observation series MSER-5 and the CI both read.

Definition at line 191 of file ldes_stats.h.

Constructor & Destructor Documentation

◆ Observations()

line::ldes::engine::Observations::Observations	(	std::size_t	M,
		std::size_t	K,
		bool	mser_on,
		int	batch )

inline

Definition at line 192 of file ldes_stats.h.

References batch_size, bt, cmp, drp, enabled, in_mser, last_qt, nclasses, nstations, qlen, and qt.

Member Function Documentation

◆ collect()

void line::ldes::engine::Observations::collect	(	const Accum &	acc,
		double	now )

inline

Record one observation.

The queue-length entry is the INTERVAL TIME-AVERAGE, not the instantaneous count: it is the integral delivered since the previous observation over the interval length, which is what makes a sequence of them comparable even though the intervals differ in duration. The other three are CUMULATIVE, because the estimators difference them against the truncation point rather than averaging them.

Definition at line 215 of file ldes_stats.h.

References bt, cmp, line::ldes::engine::Accum::completed, drp, line::ldes::engine::Accum::join_dropped, last_qt, last_time, nclasses, nstations, line::ldes::engine::Accum::qlen, qlen, qt, time, line::ldes::engine::Accum::tot_busy, and line::ldes::engine::Accum::tot_qlen.

◆ truncate()

Truncation line::ldes::engine::Observations::truncate ( ) const

inline

The truncation point, on the AGGREGATE queue length first.

The per-series fallback is not a refinement, it is the closed-network case: a closed model holds a constant total population, so the aggregate series is flat and carries no transient signal at all, however long the warmup actually was. The most conservative per-series point is taken then.

BOTH RUN OVER THE SERVICE STATIONS ONLY, which is the range applyMSER5Truncation walks. A Join is measured like a station but is not one of them, and folding its series in would move the truncation point of every fork-join model away from the reference's.

Definition at line 247 of file ldes_stats.h.

References line::ldes::engine::Truncation::applied, batch_size, enabled, in_mser, line::ldes::engine::Truncation::index, line::ldes::engine::mser5_truncation(), nclasses, nstations, qlen, time, and line::ldes::engine::Truncation::warmup_end.