The performance analysis of computer communication networks

Ji Zhang, Purdue University

Abstract

Many communication networks can be modeled as Markov chains of Quasi-Birth-Death (QBD) type, M/G/1-type or G/M/1-type. For the purpose of analyzing and optimizing the performance of these communication networks, these three classes of processes are studied extensively in this thesis. Carrier Sense Multiple Access (CSMA) networks can be modeled as QBD processes. To better understand these networks, a recently proposed matrix recursive solution is analyzed and its boundary value problem solved. It is also shown that the transient solution of QBD-processes can be expressed in a similar recursive form in the transform domain. This enables us to accurately describe the dynamic behavior of a wide class of communication networks, not just CSMA networks. Lastly, a second analytical method is proposed and compared with the matrix recursive solutions. The transient performance of ISDN networks is studied, with emphasis on the performance analysis of digital multiplexing schemes such as voice/data integrated switches. Using a novel approach, a way of computing the exact transient performance measures of a particular voice/data switch is developed. Some numerical examples are provided to show that these switches exhibit very interesting dynamic behavior. The transient behavior of time-dependent Markov processes is also studied. Such processes are used as models of networks in which the arrival or service process is time varying. A theoretical framework is laid out for the transient solution of time-dependent queues with infinite capacities. The last topic covered concerns the transient solution of semi-Markov processes used for modeling M/G/1-type queues. Such queues may prove to be necessary in the modeling of ISDN systems in which the service time distributions are arbitrary.

Degree

Ph.D.

Advisors

Coyle, Purdue University.

Subject Area

Electrical engineering|Computer science

COinS