B. Maddah ENMG 622 ENMG /27/07

Similar documents
B. Maddah INDE 504 Simulation 09/02/17

Queuing Theory. Basic properties, Markovian models, Networks of queues, General service time distributions, Finite source models, Multiserver queues

Simulation of Discrete Event Systems

First come, first served (FCFS) Batch

Announcements. Queueing Systems: Lecture 1. Lecture Outline. Topics in Queueing Theory

A queueing system can be described as customers arriving for service, waiting for service if it is not immediate, and if having waited for service,

B. Maddah ENMG 622 ENMG /20/09

Reliability and Queueing

TUFTS UNIVERSITY DEPARTMENT OF CIVIL AND ENVIRONMENTAL ENGINEERING ES 152 ENGINEERING SYSTEMS Spring Lesson 14. Queue System Theory

CS/ECE 715 Spring 2004 Homework 5 (Due date: March 16)


TCOM 501: Networking Theory & Fundamentals. Lecture 3 January 29, 2003 Prof. Yannis A. Korilis

SINGLE-CHANNEL QUEUING PROBLEMS APPROACH

Control chart for number of customers in the system of M [X] / M / 1 Queueing system

Queueing Theory (Part 3)

On Poisson Bulk Arrival Queue: M / M /2/ N with. Balking, Reneging and Heterogeneous servers

The statistical pattern of the arrival can be indicated through the probability distribution of the number of the arrivals in an interval.

1 Introduction to reducing variance in Monte Carlo simulations

Generalized Semi- Markov Processes (GSMP)

Queueing theory and Replacement model

Birth-Death Processes. Outline. EEC 686/785 Modeling & Performance Evaluation of Computer Systems. Relationship Among Stochastic Processes.

4.6 M/M/s/s Loss Queueing Model

Analysis of a queueing model with service threshold

Increasing timing capacity using packet coloring

c. Explain the basic Newsvendor model. Why is it useful for SC models? e. What additional research do you believe will be helpful in this area?

Sequences A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence

Chapter 11 Output Analysis for a Single Model. Banks, Carson, Nelson & Nicol Discrete-Event System Simulation

7.1 Convergence of sequences of random variables

Output Analysis (2, Chapters 10 &11 Law)

Empirical Distributions

A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence

FLUID LIMIT FOR CUMULATIVE IDLE TIME IN MULTIPHASE QUEUES. Akademijos 4, LT-08663, Vilnius, LITHUANIA 1,2 Vilnius University

Limit Theorems. Convergence in Probability. Let X be the number of heads observed in n tosses. Then, E[X] = np and Var[X] = np(1-p).

Introduction to probability Stochastic Process Queuing systems. TELE4642: Week2

There is no straightforward approach for choosing the warmup period l.

CEE 522 Autumn Uncertainty Concepts for Geotechnical Engineering

CHAPTER 10 INFINITE SEQUENCES AND SERIES

Problems from 9th edition of Probability and Statistical Inference by Hogg, Tanis and Zimmerman:

Infinite Sequences and Series

Queueing Theory II. Summary. M/M/1 Output process Networks of Queue Method of Stages. General Distributions

Discrete Mathematics and Probability Theory Summer 2014 James Cook Note 15

AMS570 Lecture Notes #2

A Feedback Retrial Queuing System with Starting Failures and Single Vacation

NOTES ON DISTRIBUTIONS

THE POTENTIALS METHOD FOR A CLOSED QUEUEING SYSTEM WITH HYSTERETIC STRATEGY OF THE SERVICE TIME CHANGE

Chapter 6 Sampling Distributions

Performance Evaluation of Weighted Fair Queuing System Using Matrix Geometric Method

Sojourn times in the M/G/1 FB queue with light-tailed service times

It is always the case that unions, intersections, complements, and set differences are preserved by the inverse image of a function.

Integrable Functions. { f n } is called a determining sequence for f. If f is integrable with respect to, then f d does exist as a finite real number

Application to Random Graphs

Axioms of Measure Theory

Discrete Mathematics and Probability Theory Spring 2016 Rao and Walrand Note 19

Weight Moving Average = n

6.3 Testing Series With Positive Terms

Basics of Probability Theory (for Theory of Computation courses)


ON THE RELIABILITY OF AN n-component SYSTEM 1. INTRODUCTION

Linear Regression Demystified

Econ 325/327 Notes on Sample Mean, Sample Proportion, Central Limit Theorem, Chi-square Distribution, Student s t distribution 1.

Lecture Chapter 6: Convergence of Random Sequences

Lecture 27. Capacity of additive Gaussian noise channel and the sphere packing bound

6 Integers Modulo n. integer k can be written as k = qn + r, with q,r, 0 r b. So any integer.

Sequences and Series of Functions

3. Z Transform. Recall that the Fourier transform (FT) of a DT signal xn [ ] is ( ) [ ] = In order for the FT to exist in the finite magnitude sense,

Ma 530 Introduction to Power Series

Stochastic Matrices in a Finite Field

Analysis and Approximation of Dual Tandem Queues with Finite Buffer Capacity

Approximations and more PMFs and PDFs

Addition: Property Name Property Description Examples. a+b = b+a. a+(b+c) = (a+b)+c

62. Power series Definition 16. (Power series) Given a sequence {c n }, the series. c n x n = c 0 + c 1 x + c 2 x 2 + c 3 x 3 +

Module 1 Fundamentals in statistics

Binomial Distribution

Lecture 19: Convergence

The Boolean Ring of Intervals

Lecture 8: Convergence of transformations and law of large numbers

Lecture 5: April 17, 2013

The standard deviation of the mean

The Poisson Process *

4.3 Growth Rates of Solutions to Recurrences

M/M/1 queueing system with server start up, unreliable server and balking

Random Models. Tusheng Zhang. February 14, 2013

Markov Decision Processes

Probability and statistics: basic terms

MATH 320: Probability and Statistics 9. Estimation and Testing of Parameters. Readings: Pruim, Chapter 4

1 Convergence in Probability and the Weak Law of Large Numbers

Math 10A final exam, December 16, 2016

Advanced Stochastic Processes.

2 Markov Chain Monte Carlo Sampling

Outline Continuous-time Markov Process Poisson Process Thinning Conditioning on the Number of Events Generalizations

4. Partial Sums and the Central Limit Theorem

Massachusetts Institute of Technology

Math 155 (Lecture 3)

Ma 530 Infinite Series I

7.1 Convergence of sequences of random variables

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014.

Goodness-of-Fit Tests and Categorical Data Analysis (Devore Chapter Fourteen)

0, otherwise. EX = E(X 1 + X n ) = EX j = np and. Var(X j ) = np(1 p). Var(X) = Var(X X n ) =

Sequences. Notation. Convergence of a Sequence

The Binomial Theorem

Transcription:

B. Maddah ENMG 622 ENMG 5 3/27/7 Queueig Theory () What is a queueig system? A queueig system cosists of servers (resources) that provide service to customers (etities). A Customer requestig service will start service if the required server is ot busy. Otherwise, the customer waits i queue util the server is available. Queueig (waitig i lie) happes because there are ot eough resources at certai times. µ 2 : c

Compoets of a queueig System A queuig system ca be composed of oe or may service ceters or odes. Customers are routed from oe ode to the other accordig to certai rules. Each ode is characterized by three compoets. (i) The arrival process ; (ii) The service process ; (iii) The queue disciplie. The arrival process is specified through the radom variables A, A 2,, where A i is the iter-arrival time betwee the (i ) st ad the i th customer. A typical modelig assumptio is to assume that A i s are idepedet ad idetically distributed (iid). The, the arrival process is characterized by F A (x) = P{A < x}, the cdf of A. Importat parameters of the arrival process (i additio to F A (.)) are the mea iter-arrival time E[A], ad the arrival rate = /E[A], the arrival rate. The most commoly assumed arrival process is the Poisso process. This assumptio is realistic (i most cases). I additio, it greatly simplifies the aalysis. 2

The service process is specified through the radom variables S, S 2,, where S i is the service time the i th customer. The S i are also typically assumed iid with cdf F S (x). Importat parameters of the arrival process are the mea service time E[S] ad the service rate E[S] = /µ. Service times are also commoly assumed to be expoetial. Aalytical methods that aalyze queues are quite complex without the expoetial assumptio. The queueig or service disciplie refers to the rule utilized to select the ext customer from the queue whe a customer fiishes service. Typical queueig disciplie iclude first-i, first-out (FIFO), last-i, first out (LIFO), processor sharig (PS), service i radom order (SIRO), ad priority (PR). Uder the iid assumptios, a sigle-ode queueig system is geerally deoted by GI/GI/c, where the GI refers to iid arrival ad service processes ad c is the umber of servers. If the iter-arrival ad service times are iid expoetial the the queue is deoted by M/M/c, where the M refers to the Markovia or memoryless property of the expoetial distributio. 3

Performace measures ad geeral relatios Cosider a GI/GI/c queue (to simplify thigs). I the followig we defie steady state measures, which are statistical measures after the system has bee operatioal for a time which is large eough. A importat measure is the traffic itesity, ρ = /(cµ). If ρ, the it ca be show that the queue legth will icrease idefiitely as time passes. I stable systems, ρ <. For a sigle-server system, ρ is the mea server utilizatio. The statioary system size distributio is P = lim P{ L( t) = }, t where L(t) is the umber of customers i the system at time t. The mea umber i the system is L = P =. It ca be show that L ca be estimated differetly as L = lim t Lsds ( ). t t 4

L(s) Time, s L Time, s The mea waitig time i the system is Wi i= W = lim, where W i is the waitig time of customer i i queue plus the time the customer sped i service. Amog the most importat queueig theory results is Little s law L= W. Other measures cocer waitig i queue. The mea umber i the queue is t Lq = lim t Lq( s) ds, t where L q (s) is the umber of customers i queue at time s. L q ca be also writte as L = ( c) P. q = c 5

The mea waitig time i the queue (or the mea delay) is W q = i= lim, W q i where i Wq is the waitig time of customer i i queue plus the time the customer speds i service. Little s law implies that L q = W q. Furthermore, W ad W q are related by W = W q + /µ. Multiplyig by ad applyig Little s law we get L = L q + /µ. Here, /µ ca be see as the mea umber of busy servers. (This ca i fact be also prove by Little s law.) Note that kowig oe of the four performace measures, L, W, L q, ad W q, allows determiig the other three. The M/M/ queue Cosider a sigle-server queue with iid expoetial iterarrival ad service times (hece called M/M/). Let ad µ deote the arrival ad service rates ad ρ = /µ. Assume ρ <. µ 6

The umber of customers i the M/M/ system L(t) is a birth death process with i =, ad µ i =µ, with the followig trasitio probability diagram: µ µ 2 3 µ Recall that ρ = / µ is the traffic itesity. ρ ca be also see as the average server utilizatio, or the fractio of time the server is busy. Applyig the geeral flow balace equatio for a birth-death process, the limitig probabilities are give by P 2 = + = + = + ρ = µµ µµ 2 = µ = ρ ρ ρ ρ. = ρ = + = + = = = ( ), =, 2, 2 P P ρ ρ µµ µµ 2 Note that ρ < oly if ρ < or equivaletly < µ. = This is the stability coditio that should be always satisfies i order for the M/M/ queue to have a fiite cogestio level (measured i L, W, W q, etc). 7

P could be foud i aother way by otig that P = P{server is idle} = P{server is busy} = ρ. The mea umber i the system, L, is ρ L= P = ρ ( ρ) = =. ρ µ = = Other performace measures are determied as follows: L q 2 2 ρ = L ρ = =, ρ µ ( µ ) L W = =,(Little's law) µ W q Lq = =,(Little's law) µ ( µ ) Aother measure of performace is the probability that the umber of customers i the system is or more P = lim P{ L() t } = P = ρ ( ρ)= ρ. + t k= k= Example Customers arrive at a bak accordig to a Poisso process with rate 9 customers per hour ad request service of a sigle teller. The teller has a expoetial service time with rate customers per hour. What is the fractio of time the teller is busy? This ca be modeled as a M/M/ with = 9 customers/hour ad µ = customers/hour. 8

The traffic itesity is ρ = 9/ =.9, which is also the fractio of time the server is busy. What is the mea umber of customers i the bak? L = ρ / ( ρ) =.9/. = 9 customers. What is the mea umber of customers waitig i lie? L q = L ρ = 9.9 = 8. customers. What is the mea time a customer speds i the bak? W = L / = 9 / 9 = hour. What is the mea delay i queue? W q = L q / = 8. / 9 =.9 hour = 54 mis. For what fractio of time the umber of customers i the system exceeds 3? P 4+ = ρ 4 =.9 4 =.656. What do you thik of this system performace? Not good. Mea delay time is too log. How would you improve the performace? Add oe or more servers, or trai the teller (if possible) so she hadles customers fasters (i.e. icrease µ). What other measures of performace would you estimate? P{waitig time > t } < α, where α is small. 9

Beware of the oliear behavior of queues! 5 4 L( ρ) 3 2..2.3.4.5.6.7.8.9 ρ For example, if ρ is icreased from.9 to.945 (by 5%) i the bak example, L icreases from 9 to 7.8 (by about %).