Result of pfqn_manjunath. More...

#include <line/api/pfqn/pfqn_manjunath.h>

Collaboration diagram for line::pfqn::PfqnManjunathResult< T >:

Public Attributes
T	G
	normalizing constant in the requested arithmetic
double	lG = 0.0
	log of the constant, always a double
std::size_t	peak_states = 0
	peak live series coefficients, the realised cost
bool	has_stats = false
	whether the decomposition below was computed
std::vector< double >	Q
	mean class r jobs at the queueing station
std::vector< double >	X
	class r cycle throughput
std::vector< double >	U
	class r utilization of the queueing station
std::vector< double >	think
	class r jobs genuinely thinking, X_r * Z_r
std::vector< double >	blocked
	class r jobs held at the delay by the constraint
std::vector< double >	delay
	class r jobs at the delay, think + blocked

Detailed Description

template<class T>
struct line::pfqn::PfqnManjunathResult< T >

Result of pfqn_manjunath.

The decomposition fields are populated only when PfqnManjunathOptions::stats is set. They are double in every arithmetic, exactly as lG is, because each is a RATIO of normalizing constants taken in the log domain – which is what keeps the four codebases comparable digit for digit and what removes the range problem from a ratio of two very large constants. G itself stays exact under Arith::Exact.

Q + think + blocked == N exactly: a refused admission is a DELETED transition, so a blocked job never leaves the delay, and because the think time is exponential a held job is indistinguishable from one still thinking. Little's law is what separates the two.

Definition at line 130 of file pfqn_manjunath.h.