A Structural Approach to Computer Network Simulation
Original Publication Date: 1975-Dec-31
Included in the Prior Art Database: 2005-Sep-16
Publishing Venue
Software Patent Institute
Related People
G. Michael Schneider: AUTHOR [+3]
Abstract
Much of the initial development work in the area of computer networks has concentrated on the design, of the fundamental communication subnetwork and associated protocols necessary to support such a network. This would include such areas as: a) Control Responsibility -- How many nodes within the network should have communication control responsibilities? Within any individual node where should that responsibility lie? b) Network Topology and Optimization -- Using the specified control structure and available communication links, how can we interconnect all network nodes so as to minimize costs and delays while maximizing throughput? c) Reliability -- A topology that is optimal with respect to costs and delay times must still minimize the probability of a working node becoming disconnected from the remainder of the network. d) Subnetwork Protocols -- What are the "optimal" protocols for supporting such diverse subnetwork functions as routing, error detection, error correction, and flow control? One of the primary tools for investigating the above questions has been the use of discrete simulation models. However, the models reported in the literature have usually been limited in scope -- being confined to the investigation of a single aspect of the overall network design problem. Examples of this approach might include models that do a performance evaluation on distributed networks jl], heuristic cost optimization studies [2], response time simulation of terminal-concentrator complexes [13], or models of specific existing computer networks [3;4,12,14]. Other simulation models are discussed and described in [5;6,15,16,17]. The above are all examples of what we term parametric simulation models --models in which the structure of the subnetwork protocols are fixed (or ignored) by the model and are invariant:. The only control a user exercises over the model is via the inputting of a small amount of parametric data --typically such values as number of nodes, internodal distances, line speeds, message frequencies, message lengths, and channel error rates. Higher level subnetwork functions such as error detraction and correction procedures, routing algorithms, achnowledgement procedures or connection protocols are usually not part of the user/model interface and are changed only by major modifications to the model itself. While these parametric models usually give good results in their specific areas of investigation they are deficient in a number of ways: a) Their inflexible assumptions about the subnetwork organization frequently makes the model not easily applicable to different networks with different architectures (e.g. a message-switched vs. a circuit-switched connection protocol). b) A model useful for studying one :aspect of the network problem is frequently inappropriate for investigating a completely different problem -- thus necessitating multiple models of the same network. In [16J Chou describes 6 separate? independent programs that would be needed to study various stages in the network design process. c) Much of the current research in network design is going beyond simple parametric investigations or sensitivity analysis and, instead, is focusing on such topics as multiprocessor communications handlers [7], local and global dynamic routing algorithms [8], internetwork interfaces [9], and the use of satellite communication technology [10].
THIS DOCUMENT IS AN APPROXIMATE REPRESENTATION OF THE ORIGINAL.
A Structural Approach to Computer Network Simulation
by
G. Michael Schneider
Computer, Information, and Control Sciences Department
114 Lind Hall
Institute of Technology
University of Minnesota
Minneapolis, Minnesota 55455
Technical Report 75-20
December, 1975 The Network Simulation Project Working Paper No. 2
Cover design courtesy of Ruth and Jay Leavitt A Structural. Approach to Computer Network Simulation
I. Introduction
Much of the initial development work in the area of computer networks has concentrated on the design, of the fundamental communication subnetwork and associated protocols necessary to support such a network. This would include such areas as: a) Control Responsibility -- How many nodes within the network should have communication control responsibilities? Within any individual node where should that responsibility lie? b) Network Topology and Optimization -- Using the specified control structure and available communication links, how can we interconnect all network nodes so as to minimize costs and delays while maximizing throughput? c) Reliability -- A topology that is optimal with respect to costs and delay times must still minimize the probability of a working node becoming disconnected from the remainder of the network. d) Subnetwork Protocols -- What are the "optimal" protocols for supporting such diverse subnetwork functions as routing, error detection, error correction, and flow control? One of the primary tools for investigating the above questions has been the use of discrete simulation models. However, the models reported in the literature have usually been limited in scope -- being confined to the investigation of a single aspect of the overall network design problem. Examples of this approach might include models that do a performance evaluation on distributed networks jl], heuristic cost optimization studies [2], response time simulation of terminal-concentrator complexes [13], or models of specific existing computer networks [3;4,12,14]. Other simulation models are discussed and described in [5;6,15,16,17]. The above are all examples of what we term parametric simulation models --models in which the structure of the subnetwork protocols are fixed (or ignored) by the model and are invariant:. The only control a user exercises over the model is via the inputting of a small amount of parametric data --typically such values as number of nodes, internodal distances, line speeds, message
University of Minnesota Page 1 Dec 31, 1975
A Structural Approach to Computer Network Simulation
frequencies, message lengths, and channel error rates. Higher level subnetwork functions such as error detraction and correction procedures, routing algorithms, achnowledgement procedures or connection protocols are usually not part of the user/model interface and are changed only by major modifications to the model itself. While these parametric mode...