J.

MEDHI

STOCHASTIC
MODELS IN
QUEUEING THEORY
SECOND EDITION
ACADEMIC PRESS
An imprint of Elsevier Science

Amsterdam Boston London New York Oxford Paris

San Diego San Francisco Singapore Sydney Tokyo


Contents

Preface xv

CHAPTER 1 Stochastic Processes 1

1.1 Introduction 1
1.2 Markov Chains 2
1.2.1 Basic ideas 2
1.2.2 Classification of states and chains 4
1.3 Continuous-Time Markov Chains 14
1.3.1 Sojourn time 14
1.3.2 Transition density matrix or infinitesimal
generator 15
1.3.3 Limiting behavior: ergodicity 16
1.3.4 Transient solution 18
1.3.5 Alternative definition 19
1.4 Birth-and-Death Processes 23
1.4.1 Special case: M/M/1 queue 25
1.4.1 Pure birth process: Yule-Furry process 25
1.5 Poisson Process 25
1.5.1 Properties of the Poisson process 28
1.5.2 Generalization of the Poisson process 29
1.5.3 Role of the Poisson process in probability
models 31
1.6 Randomization: Derived Markov Chains 32
1.6.1 Markov chain an an underlying Poisson process
(or subordinated to a Poisson process) 33
viii Contents

1.6.2 Equivalence of the two limiting forms 33

1.6.3 Numerical method 34
1.7 Renewal Processes 35
1.7.1 Introduction 35
1.7.2 Residual and excess lifetimes 36
1.8 Regenerative Processes 37
1.8.1 Application in queueing theory 38
1.9 Markov Renewal Processes and Semi-Markov
Processes 39
Problems 41
References and Further Reading 46
CHAPTER 2 Queueing Systems: General Concepts 47
2.1 Introduction 47
2.1.1 Basic characteristics 48
2.1.2 The Input or arrival pattern of customers 48
2.1.3 The pattern of service 49
2.1.4 The number of servers 49
2.1.5 The capacity of the system 49
2.1.6 The queue discipline 49
2.2 Queueing Processes 50
2.3 Notation 51
2.4 Transient and Steady-State Behavior 52
2.5 Limitations of the Steady-State Distribution 53
2.6 Some General Relationships in Queueing Theory 54
2.7 Poisson Arrival Process and Its Characteristics 59
2.7.1 PASTA: Poisson arrivals see time averages 59
2.7.2 ASTA: arrivals see time averages 62
References and Further Reading 62
CHAPTER 3 Birth-and-Death Queueing Systems:
Exponential Models 65
3.1 Introduction 65
3.2 The Simple M/11111 Queue 65
3.2.1 Steady-state solution of M/A4/1 66
3.2.2 Waiting-time distributions 68
3.2.3 The output process 72
3.2.4 Semi-Markov process analysis 75
3.3 System with Limited Waiting Space:
The MIM11/K Model 77
3.3.1 Steady-state solution 77
3.3.2 Expected number in the system LK 78
3.3.3 Equivalence of an M/M/1/K model with
a two-stage cyclic model 80


Contents ix

3.4 Birth-and-Death Processes: Exponential Models 81

3.5 The MPti/o0 Model: Exponential Model with an Infinite
Number of Servers 83
3.6 The Model M/M/c 84
3.6.1 Steady-state distribution 84
3.6.2 Expected number of busy and idle servers 87
3.6.3 Waiting-time distributions 89
3.6.4 The output process 93
3.7 The M/M/c/c System: Erlang Loss Model 95
3.7.1 Erlang loss (blocking) formula:
Recursive algorithm 99
3.7.2 Relation between Erlang's B and Cformulas 100
3.8 Model with Finite Input Source 101
3.8.1 Steady-state distribution: M/M /c/ Irn(m>c).
Engset delay model 101
3.8.2 Engset loss model MIMIcIlm/(rn > c) 106
3.8.3 The mode! M/M /c //m (m < c) 109
3.9 Transient Behavior 110
3.9.1 Introduction 110
3.9.2 Difference-equation technique 112
3.9.3 Method of generating function 117
3.9.4 Busy-period analysis 119
3.9.5 Waiting-time process: Virtual waitingtime 125
3.10 Transient-State Distribution of the M/M/c Model 127
3.10.1 Solution of the differential-difference
equations 127
3.10.2 Busy period of an M /M/c queue 133
3.10.3 Transient-state distribution of the output
of an M /M /c queue 136
3.11 Multichannel Queue with Ordered Entry 138
3.11.1 Two-channel model with ordered entry
(with finite capacity) 139
3.11.2 The case M 1, N=N 140
3.11.3 Particular case: M N 1 (overflow system) 142
3.11.4 Output process 144
Problems and Complements 145
References and Further Reading 159

CHAPTER 4 Non- Birth -and- Death Queueing Systems:

Markovian Models 165
4.1 Introduction, 165
4.1.1 The system M/Ek/1 165
4.1.2 The system Ek / M/1 170


X Contents

4.2 Bulk Queues 174

4.2.1 Markovian bulk-arrival system: WM/1 174
4.2.2 Equivalence of Ms. /M/1 and M/E,./1
systems 178
4.2.3 Waiting-time distribution in an WM/1
queue 178
4.2.4 Transient-state behavior 179
4.2.5 The system MVM/00 181
4.3 Queueing Models with Bulk (Batch) Service 185
4.3.1 The system M/M(a, b)/1 186
4.3.2 Distribution of the waiting-time for the system
M/M(a, b)/1 190
4.3.3 Service batch-size distribution 195
4.4 M/M(a, b)/1: Transient-State Distribution 196
4.4.1 Steady-state solution 198
4.4.2 Busy-period distribution 198
4.5 Two-Server Model: M/M(a, b)/2 202
4.5.1 Particular case: M/M(1, b)/2 204
4.6 The M /M(1, b)/c Model 205
4.6.1 Steady-state resultsM/M(1, b)/c 208
Problems and Complements 210
References and Further Reading 217
CHAPTER 5 Network of Queues 221
5.1 Network of Markovian Queues 221
5.2 Channels in Series or Tandem Queues 222
5.2.1 Queues in series with multiple channels at each
phase 224
5.3 Jackson Network 226
5.4 Closed Markovian Network
(Gordon and Newell Network) 233
5.5 Cyclic Queue 236
5.6 BCMP Networks 238
5.7 Concluding Remarks 240
5.7.1 Loss networks 241
Problems and Complements 242
References and Further Reading 249

CHAPTER 6 Non - Markovian Queueing Systems 255

6.1 Introduction 255
6.2 Embedded-Markov-Chain Technique
for the System with Poisson Input 256


Contents Xi
6.3 The MIGIl Model: Pollaczek-Khinchin Formula 259
6.3.1 Steady-state distribution of departure
epoch system size 259
6.3.2 Waiting-time distribution 261
6.3.3 General time system size distribution
of an M/G/1 queue: supplementary
variable technique 267
6.3.4 Semi-Markov process approach 274
6.3.5 Approach via martingale 274
6.4 Busy Period 276
6.4.1 Introduction 276
6.4.2 Busy-period distribution: Taketcs integral
equation 277
6.4.3 Further discussion of the busy period 279
6.4.4 Delay busy period 284
6.4.5 Delay busy period under N-policy 285
6.5 Queues with Finite Input Source:
MIG111 IN System 289
6.6 System with Limited Waiting Space:
M/G/1/K System 292

6.7 The M7G/1 Model with Bulk Arrival 295

6.7.1 The number in the system at departure
epochs in steady state (Pollaczek-Khinchin
formula) 295
6.7.2 Waiting-time distribution 295
6.7.3 Feedback queues 302
6.8 The M /et, b)/1 Model with General
Bulk Service 304
6.9 The G/M/1 Model 306
6.9.1 Steady-state arrival epoch system size 306
6.9.2 General time system size
in steady state 309
6.9.3 Waiting-time distribution 311
6.9.4 Expected duration of busy period
and idle period 313
6.10 Multiserver Model 314
6.10.1 The M/G/00 model: transient-state
distribution 314
6.10.2 The model GIMIc 319
6.10.3 The model M/G/c 322
6.11 Queues with Markovian Arrival Process 324
Problems and Complements 326
References and Further Reading 334
Xii Contents

CHAPTER 7 Queueswith General Arrival Time and Service-Time

Distributions 339
7.1 The G/G/1 Queue with General Arrival Time
and Service-Time Distributions 339
7.1.1 Lindley's integral equation 341
7.1.2 Laplace transform of VJ 343
7.1.3 Generalization of the Pollaczek-Khinchin
transform formula 346
7.2 Mean and Variance of Waiting Time W 348
7.2.1 Mean of W (single-server queue) 348
7.2.2 Variance of IN 351
7.2.3 Multiserver queues: approximation of mean
waiting time 353
7.3 Queues with Batch Arrivals G('`) IGI1 356
7.4 The Output Process of a G/G/1 System 358
7.4.1 Particular case 359
7.4.2 Output process of a G/G/c system 360
7.5 Some Bounds for the G/G/1 System 360
7.5.1 Bound for e) 360
7.5.2 Bounds for E(W) 360
Problems and Complements 368
References and Further Reading 371

CHAPTER 8 Miscellaneous Topics 375

8.1 Heavy-Traffic Approximation for Waiting-Time
Distribution 375
8.1.1 Kingman's heavy-traffic approximation
for a G/G/1 queue 375
8.1.2 Empirical extension of the AVG/1 heavy-traffic
approximation 379
8.1.3 GIM1c queueinheavy traffic 381
8.2 Brownian Motion Process 383
8.2.1 Introduction 383
8.2.2 Asymptotic queue-length distribution 386
8.2.3 Diffusion approximation for
a G/G/1 queue 389
8.2.4 Virtual delay for th'e G/G/1 system 391
8.2.5 Approach through an absorbing barrier
with instantaneous return 394
8.2.6 Diffusion approximation for a G/G /c queue:
state-`dependent diffusion equation 395


Contents xiii

8.2.7 Diffusion approximation for an M/G/c model 396

8.2.8 Concluding remarks 397
8.3 Queueing Systems with Vacations 398
8.3.1 Introduction 398
8.3.2 Stochastic decomposition 399
8.3.3 Poisson Input queue with vacations:
[exhaustive-service] queue-Iength
distribution 399
8.3.4 Poisson input queue with vacations:
waiting-time distribution 404
8.3.5 M/G/1 system with vacations:
nonexhaustive service 406
8.3.6 Limited service system: MIG11-1/a, model 407
8.3.7 Gated service system: MIG11-14 model 408
8.3.8 MIG111K queue with multiple vacations 412
8.3.9 Mean value analysis through heuristic
treatment 416
8.4 Design and Control of Queues 423
8.5 Retrial Queueing System 427
8.5.1 Retrial queues: model description 427
8.5.2 Single-server model: MIM /1 retrial queue 429
8.5.3 M/G/1 retrial queue 432
8.5.4 Multiserver models 436
8.5.5 Model with finite orbit size 439
8.5.6 Other retrial queue models 440
8.6 Emergence of a New Trend in Teletraffic Theory 441
8.6.1 Introduction 441
8.6.2 Heavy-tail distributions 442
8.6.3 M/G/1 with heavy-tailed service time 445
8.6.4 Pareto mixture of exponential (PME)
distribution 445
8.6.5 Gamma mixture of Pareto (GMP) distribution 447
8.6.6 Beta mixture of exponential (BME)
distribution 450
8.6.7 A dass of heavy-tail distributions 452
8.6.8 Long-range dependence 454
Problems and Complements 455
References and Further Reading 461
Appendix 469
Index 477