ON SOME DESIGN ASPECTS OF COMPUTER COMMUNICATION NETWORKS

Abstract

Computer communication network (CCN) have evolved very rapidly during the last decade. Recently, as a part of continuing advancement in the networking technology, several impressive developments have been witnessed in the form of broadcast networks, local area networks, and integrated voice and data networks. ARPANET (Advanced Research project Agency), the first major network, has generated tremendous interest in the field of computer communication network. Today CCN of various kinds are currently in existence or in the process of being set up. These include ALOHA (developed at University of Hawaii), aUTOVON (automatic Voice Network), commercial network architectures SNA and DNA (developed by IBM and DEC res pectively), and public data networks TYMNET and TRANSPAC (developed by Tymshare and PTT of French Government).The number is endless and growing larger literally day by day. The major design problems in computer communication network include topological optimization for cost, delay and throughput j capacity assignment to links, routing techniques, network reliability, node to node call blocking and so on. Out of these, the last four problems are studied in detail in chapters 3 thru' 5 of this disser tation. \ For the reliability analysis, it is generally assumed that requisite amount of information is always transmitted

from source to destination whenever a path is available between them. This implication is neither valid nor eco nomically justifiable in the design of aCCN. In practice, a communication link has a limited capacity because of the cost of aCCN being mainly dependent on it. In chapter 3, we consider a design problem and define the integration of the link capacity with its reliability as a measure of the performance index of a CCN. A method is proposed to solve this problem. It uses the concept of terminal relia bility evaluation. The method is simple and advantageous as it applies even for a large network. The routing issue is the problem of establishing a conti nuous path, usually incorporating several links in a net work, between any pair of source and destination node along which message is to be sent. The objective of any routing procedure is to obtain a minimum time delay while maintai ning high throughput. Chapter 4 of the thesis deals with this problem. An algorithm is given to enumerate all the paths between any node acting as source node to another node acting as terminating node, along which message can be routed. The method utilizes the concept of simple algebra. The method is straightforward, easy to computerize and computationally efficient. The concept of path enumeration is, further, exten ded to obtain the path-loss sequence in various routing plans such as successive-office control (SOC), originatingoffice control (00C) and 00C with spill- forward. These

strategies are quite practical. The European AUTOVON network employs 00C with spill-forward and C0NU3 (CONtinental ug). uses SOC. As against usual augmented route: tree approach for obtaining path- loss sequence, the proposed method is simple and can be implemented easily over a general purpose computer. In a circuit— switched network one important measure of quality of service is node to node grade of service (NNGOS) which is blocking probability of calls originated at source node and destined for terminal node. NNGOS is the totality of conditions under which a call at source node is not allowed to reach terminal node. This congestion value is mainly de pendent on link blocking probabilities, topological layout of network and call control strategy considered above. Chapter 5 of the thesis discusses this problem for .its various relevant aspects. Path- loss sequence enumerated in chapter 4 is consi dered as input and utilizing an efficient reliability evalua tion technique the NNGOS in a circuit- switched network is computed. Note the two-step procedure uses the path-loss sequence information' which, in turn, depends on routing stra tegy. The complexity involved in enumerating path- loss sequ ence may be large in some CCNs. Therefore, another method which does not require the apriori knowledge of path-loss sequence is also proposed. The method is recursive and is derived from link by link call set up procedure. Moreover,it is a particular version of decomposition approach in which a succession of key- stone element(s) is choosen. The method applies to network having unreliable nodes des cribed in terms of input and output switch blocking probabilities. The advantage with this method is that it requires lesser number of steps as compared to exis ting methods for computing node to node congestion in a circuit- switched network and is also applicable to arbitrary routing plan (OOC with spill) besides being applicable for SOC and OOC strategies. The algorithm re sults symbolic reliability expression for a general net work under minor modifications discussed in the text. Chapter 6 finally concludes the dissertation. It also pre sents a number of problems as a future scope of the present work.