182 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL 41, NO. 2, APRIL 1994 A Remote Bridging Technique to Increase Performability in Distributed Systems S, Casale, Member, IEEE, V. Catania, A. Puliaito, and . Vita, Member, IEEE Absrct- The paper presents a network architecture and protocol to extend the facilities available on LAN's to a wider area. The proposal, based on a remote briding technique, al lows selective network reconiguration and consequently high network availability and performability values. This make the system suitable for applications in environments characterized by hard reliability constraints. The proposed solution is assssed by modeling it through Markov chain hoy. I. INTRODUCTION I NTERCONNECTION of LAN's by means of MAC bridges is a technique which has been used more and more to allow sharing of data facilities in customer local premises. The extension of this technique to geographically remote areas has today become a challenge for broadband network technologies. In fact, with increasing frequency, many users require the networking of their LAN's, spread in wide areas, to extend the facilities already available on single LAN's, mantaining high levels of performance and reliahility. In this context, the adoption of routing schemes which usually rely on bridges seems to be a very attractive idea [3], [6], [7]. In this paper we focus on the spanning tree CST) algorithm [I], which is one of the most popular routing schemes used by bridges. This scheme imposes some relevant constraints on the size of the network, mainly due to the maximum frame forwarding delay and network availability requirements, as deined in [11]. Both of these requirements set a limit on the growth of a bridged LAN. As far as frame forwarding delay is concened, the constraint can largerly he removed thanks to the high bit rate which is now available on high speed networks. By using these technologies as the interconnection network between bridged LAN's, it is in fact possible to obtain extended networks with low forwarding delay values [SJ. Meeting network availability requirements depends, on the other hand, on the network reconiguration processes. In the event of failures or variations in topology, the ST algorithm allows the network to be automatically reconigured, the ecovery time being proportional to the topology diameter [17]. During this phase, due to the presence of possible multiple paths, no frames are routed in the involved network sctions or, in some cases, in the whole network, thus leading to evice unavailability. Furthermore, once the network has een reconfigured, there is a temporary increase in trafic, namely broadcast traic, as bridge forwarding tables have to Manuscript received November 30, 1992; revised March 5, 1993. The authos re wi th the Isrituto di Informatica e Telecomunicazioni Fcola' di Ingcneria-Universita' di Catania, 95125 Catania, Italy. ' lEEE Log Number 9215731. be updated, with a consequent degradation in the quality of service. Obviously, the limits mentioned above weigh more heavily if the size of the network increases, because when the number of components in a bridged LAN (LAN's and bridges) grows, there is a corresponding increase in the occurrence of causes requiring reconiguration and the time it takes to reconigure the network. We address these issues proposing an enhanced ST algorithm which scales up eficiently to large network sizes. Some relevant works [9J, [14J have dealt with the prob lem of extcnding the ST to wide area networks. In [9], 1. Hart proposes extending thc ST standard to remote bridges operating on backbone networks. A hypothesis which allows this coniguration to work in a consistent and deterministic way is that all local and remote bridges support the same ST algorithm. In [i4J W . D. Sinkoskie and C. 1. Cotton propose a scheme of adaptive routing for a nation wide network based on logical decomposition of the network into a multiple ST which even works in the presence of loops in the active topology. The work focuses on the reduction of multicast traic and on automatic network coniguration to follow movements of users. In our proposal we define a routing scheme which aims at improving network availability through a drastic reduction in reconiguration times, especially in large networks. We assume that the network is made up of autonomous bridged LAN's (islands), each containing local bridges [i] operating according to the standard, interconnected by remote bridges. The routing strategy is based on a two-level spanning tree: the lower level, supported by local bridges, and the upper level supported by remote bridges. This kind of organization allows the network to be reconfigured selectively, with the advantages of both service continuity and a notahle reduction in the average amount of traffic lost due to reconiguration processes. The paper is organized into the following sections. Section II presents a survey of the spanning tree algorithm and protocol; Section III deals with the two levels spanning tree solution and presents the interconnection devices it is based on; Section IV gives some remarks on evaluation concening our proposal; Section V presents an analytical model of the network and analyzes some performance and reliability parameters, namely network availability and performability; Section VI presents our conclusions. n, A SURVEYOr THE SPANNING TREE PROTOCOL A MAC bridge [I], [10], [II] is made up of a set of ports, each of which is connected to an individual LAN, and a MAC rclay entity, which has two main functions (Fig. 1): 0278-046/94$04.00 @ 1994 IEEE