Properties and Bounds on P/T NetsJavier CamposUniversidad de Zaragoza (Spain)Tutorial of PNPM’99 – PAPM’99 – NSMC’99Zaragoza (Spain), September 6-10, 1999Preliminary comments (1)• Interest of bounding techniques– preliminary phases of design• many parametersexactare not known accuracysolutionaccurately• quick evaluation andboundsrejection of thoseclearly badcomplexityProperties and Bounds on P/T Nets Javier CamposTutorial of PNPM’99 – PAPM’99 – NSMC’99, Zaragoza (Spain), Sep. 6-10, 1999 p. 2Preliminary comments (2)• Net-driven solution technique– stressing the intimate relationship betweenqualitative and quantitative aspects of PN’s– structure theory of net modelsefficient computation techniquesProperties and Bounds on P/T Nets Javier CamposTutorial of PNPM’99 – PAPM’99 – NSMC’99, Zaragoza (Spain), Sep. 6-10, 1999 p. 3Outline• Introducing the ideas: Marked Graphs case• Generalization: use of visit ratios• Improvements of the bounds• A general linear programming statementProperties and Bounds on P/T Nets Javier CamposTutorial of PNPM’99 – PAPM’99 – NSMC’99, Zaragoza (Spain), Sep. 6-10, 1999 p. 4Introducing ideas: MG’s case (1)t2 p4p2p1 t1 t4 generally distributed service times (random variables X with mean ) s [t ] it3 j p5p3 we assume infinite-server semanticsexact cycle time (random variable): X = X + max{X , X }+ X 1 2 3 4 average cycle time: G = s [t ]+ E[max{X , X }]+ s [t ] 1 2 3 4 but (non-negative variables): X ...
Properties and Bounds on P/T Nets Javier Campos Universidad de Zaragoza (Spain)
TutorialofPNPM’99PAPM’99NSMC’99
Zaragoza (Spain), September 6-10, 1999
Preliminary comments (1)
• Interest of bounding techniques preliminary phases of design • many parameters are not known accurately • quick evaluation and rejection of those clearly bad
accuracy
bounds
Properties and Bounds on P/T Nets TutorialofPNPM’99PAPM’99NSMC’99,Zaragoza(Spain),Sep.6-10,1999
exact solution
complexity
Javier Campos p. 2
•
Preliminary comments (2)
Net-driven solution technique stressing the intimate relationship between qualitative and quantitative aspects of PN’s structure theory of net models
efficient computation techniques
Properties and Bounds on P/T Nets Javier Campos TutorialofPNPM’99PAPM’99NSMC’99,Zaragoza(Spain),Sep.6-10,1999p.3
• • • •
Outline
Introducing the ideas: Marked Graphs case Generalization: use of visit ratios Improvements of the bounds A general linear programming statement
Properties and Bounds on P/T Nets TutorialofPNPM’99PAPM’99NSMC’99,Zaragoza(Spain),Sep.6-10,1999
Javier Campos p. 4
Introducingideas:MG’scase(1)
p1 t1 p2
p3
t2 t3
p4 t4generally distributed service times (random variablesXiwithnmaes[tj]) p5we assumeinfinite-server semantics
exact cycle time (random variable):X=X1+max{X2,X3}+X4average cycle time:G =s[t1]+E[max{X2,X3}]+s[t4] but (non-negative variables):X2,X3£max{X2,X3}£X2+X3therefore:s[t1]+max{s[t2],s[t3]}+s[t4] ££ Gs[t1]+s[t2]+s[t3]+s[t4]
Properties and Bounds on P/T Nets Javier Campos TutorialofPNPM’99PAPM’99NSMC’99,Zaragoza(Spain),Sep.6-10,1999p.5
Introducingideas:MG’scase(2)
Thus, the lower bound for the average cycle time is computed looking for the slowest circuit
æås[ti]ö G ³maxçtiÎC÷CÎ{circuitsç in# tokensC÷ of the net}è ø
Interpretation: an MG may be built synchronising circuits, so we look for the bottleneck
Properties and Bounds on P/T Nets Javier Campos TutorialofPNPM’99PAPM’99NSMC’99,Zaragoza(Spain),Sep.6-10,1999p.6
Introducingideas:MG’scase(3)
(sis the vector of average service times)
• Computation:G ³maximumy×Pre×s subject toy×C=0 y×m0=1 y³0 (the proof of this comes later for a more general case)
solving a linear programming problem (polynomial complexityon the net size)
Properties and Bounds on P/T Nets Javier Campos TutorialofPNPM’99PAPM’99NSMC’99,Zaragoza(Spain),Sep.6-10,1999p.7
Introducingideas:MG’scase(4)
•Evenifnaïf,theboundsaretight! • Lower bound for the average cycle time
max{s[t2],s[t3]}£E[max{X2,X3}] it is exact for deterministic timing it cannot be improved using only mean values of r.v. (it is reached in a limit case for a family of random variables with arbitrary means and variances)
Properties and Bounds on P/T Nets Javier Campos TutorialofPNPM’99PAPM’99NSMC’99,Zaragoza(Spain),Sep.6-10,1999p.8
they behave “as deterministic for the ‘max’and ‘+’operators in the limit (a®1)
Properties and Bounds on P/T Nets Javier Campos TutorialofPNPM’99PAPM’99NSMC’99,Zaragoza(Spain),Sep.6-10,1999p.9
Introducingideas:MG’scase(6)
• Upper bound for the average cycle time
G £ ås[t] tÎT
it cannot be improved for 1live MG’s using only mean values of r.v. (it is reached in a limit case for a family of random variables with arbitrary means)
Properties and Bounds on P/T Nets Javier Campos TutorialofPNPM’99PAPM’99NSMC’99,Zaragoza(Spain),Sep.6-10,1999p.10