Skip to main content

1999 | Buch

Introduction to Stochastic Networks

verfasst von: Richard Serfozo

Verlag: Springer New York

Buchreihe : Stochastic Modelling and Applied Probability

insite
SUCHEN

Über dieses Buch

In a stochastic network, such as those in computer/telecommunications and manufacturing, discrete units move among a network of stations where they are processed or served. Randomness may occur in the servicing and routing of units, and there may be queueing for services. This book describes several basic stochastic network processes, beginning with Jackson networks and ending with spatial queueing systems in which units, such as cellular phones, move in a space or region where they are served. The focus is on network processes that have tractable (closed-form) expressions for the equilibrium probability distribution of the numbers of units at the stations. These distributions yield network performance parameters such as expectations of throughputs, delays, costs, and travel times. The book is intended for graduate students and researchers in engineering, science and mathematics interested in the basics of stochastic networks that have been developed over the last twenty years. Assuming a graduate course in stochastic processes without measure theory, the emphasis is on multi-dimensional Markov processes. There is also some self-contained material on point processes involving real analysis. The book also contains rather complete introductions to reversible Markov processes, Palm probabilities for stationary systems, Little laws for queueing systems and space-time Poisson processes. This material is used in describing reversible networks, waiting times at stations, travel times and space-time flows in networks. Richard Serfozo received the Ph.D. degree in Industrial Engineering and Management Sciences at Northwestern University in 1969 and is currently Professor of Industrial and Systems Engineering at Georgia Institute of Technology. Prior to that he held positions in the Boeing Company, Syracuse University, and Bell Laboratories. He has held

Inhaltsverzeichnis

Frontmatter
1. Jackson and Whittle Networks
Abstract
This chapter describes the equilibrium behavior of Jackson and Whittle networks. In such a network, the numbers of discrete units or customers at the nodes are modeled by multidimensional Markov processes. The main results characterize the equilibrium distributions of the processes. These distributions yield several performance parameters of the networks including throughput rates, expected customer waiting times, and expected busy periods for servers.
Richard Serfozo
2. Reversible Processes
Abstract
An ergodic Markov process is reversible if, in equilibrium, the expected number of transitions per unit time from one state to another is equal to the expected number of the transitions in the reverse order. This is also equivalent to a time-reversibility property that, at any instant, the future of the process is stochastically indistinguishable from viewing the process in reverse time. A remarkable feature of such a process is that its equilibrium distribution is readily obtainable as a certain product of ratios of its transition rates. A classic example is a birth-death queueing process. This chapter describes a wide class of reversible Markov network processes with batch or multiple-unit movements as well as single-unit movements. Examples include multivariate birth-death processes with single and batch increments and reversible Jackson and Whittle processes. The last two sections cover partition-reversible processes, which are generalizations of reversible processes. Invariant measures for such processes are obtainable by solving balance equations separately on subsets that partition the state space.
Richard Serfozo
3. Miscellaneous Networks
Abstract
This chapter deals with several applications and variations of the network models developed in the preceding chapters. We first show how to use Whittle processes to model networks with multiple types of units, where the routings and services may depend on a customer’s type. This includes Kelly networks with deterministic routes for units, and BCMP networks with Cox and general service times depending on a unit’s type. We also discuss several forms of blocking in networks, and bottlenecks in closed Jackson networks. The chapter ends with a discussion of partial balance equations in modeling networks.
Richard Serfozo
4. Network Flows and Travel Times
Abstract
This chapter addresses the following questions about movements of units in stationary Jackson and Whittle processes. What flows of units between nodes are Poisson processes? When a unit moves from one node to another, what is the probability distribution of the locations of the other units in the network? What is the distribution of the time it takes for a typical unit to traverse a series of nodes?
Richard Serfozo
5. Little Laws
Abstract
In a Markovian, regenerative, or stationary network, the average sojourn times of customers in a sector of the network can often be obtained from a Little law. Specifically, a Little law for a service system says that the average sojourn time W of a customer in the system and the average queue length L of the system are related by L = λW, where λ is the average arrival rate of units to the system. This fundamental relation is a law of averages or law of large numbers when the quantities L, λ, W are “limits” of averages. It is also a law of expectations when the quantities are expected values. This chapter focuses on Little laws of averages, which are based on sample path analysis. The next chapter covers Little laws of expectations for stationary systems, which are based on Palm probability analysis.
Richard Serfozo
6. Stationary Systems
Abstract
This chapter is an introduction to the basics of stationary processes and Palm probabilities that are used in queueing theory. This includes Palm calculus and Campbell—Mecke formulas for functionals of stationary systems. This material is the foundation for modeling networks and queueing systems with stationary dynamics, and for obtaining Little laws for such systems.
Richard Serfozo
7. Networks with String Transitions
Abstract
Chapter 2 discussed network models with batch or concurrent movements of units under reversibility assumptions. Are there comparable models of Whittle networks with batch movements? More generally, are there tractable network models in which a transition may involve a series of simultaneous single- or multiple-unit changes? This chapter describes networks with such characteristics called networks with string transitions, or string-nets for short. In a string-net, a transition consists of a string of instantaneous subtractions or additions of units at the nodes, where the string is randomly selected from a family of variable-length strings. Invariant measures for string-nets resemble those of Whittle networks, but now key parameters in the measures are obtained as solutions to more complicated nonlinear traffic equations.
Richard Serfozo
8. Quasi-Reversible Networks and Product Form Distributions
Abstract
This chapter addresses the following question for a Markov network process. Under what conditions is the stationary distribution of the process a product of stationary distributions associated with the nodes? We consider a network in which the state of each node may contain more information than the number of units at the node, and a network transition may be triggered by an internal node change as well as by a unit moving from one node to another. The network process is viewed as a linkage of certain artificial Markov “node processes” that mimic the operation of the nodes as if they were operating in isolation. The main results are necessary and sufficient conditions under which the stationary distribution of the network is a product of the stationary distributions of the individual node processes.
Richard Serfozo
9. Space—Time Poisson Models
Abstract
This chapter covers space—time Poisson models for queueing networks, spatial service or storage systems, and particle systems. Such a model describes the collective movement of units or customers in space and time, where the units enter the system according to a Poisson space—time process and then move about independently of each other. Because of these properties, the evolution of the system can be formulated by certain “random transformations” of Poisson point processes in space and time. We characterize these transformations and then use them in a variety of models. An important example is a network with time-dependent Poisson arrival process and infinite-server nodes with general service times.
Richard Serfozo
10. Spatial Queueing Systems
Abstract
This chapter describes a spatial queueing model for stochastic service systems in which customers or units move about and receive services in a region or a general space. The state of such a system is a point process on a space that evolves over time as a “measure-valued” Markov jump process. Each unit moves in the space according to a Markovian routing mechanism and it receives services at the locations it visits. The service times are exponentially distributed and the rates, as in a queueing system, depend on the congestion or configuration of the points in the system. The types of dependencies are extensions of those in Jackson and Whittle queueing networks.
Richard Serfozo
Backmatter
Metadaten
Titel
Introduction to Stochastic Networks
verfasst von
Richard Serfozo
Copyright-Jahr
1999
Verlag
Springer New York
Electronic ISBN
978-1-4612-1482-3
Print ISBN
978-1-4612-7160-4
DOI
https://doi.org/10.1007/978-1-4612-1482-3