Performance Improvement for a Metropolitan WDM Ring Network Architectre: An Effcient Traffc Management and a Delay-Sensitive Transmission Technique Peristera A. Baziana, Ioannis E. Pountourakis, Member, IEEE School of Electrical and Computer Engineering National Technical University of Atens Athens, Greece e-mail: baziana@central.ntua.g.ipount@cs.ntua.gr Abstact-In this paper, an efcient trafc allocation technique and a slotted delay-sensitive WDM protocol suitable for ring metropolitan area networks is proposed. Our purpose is to improve the limited bandwidth utilization that many WDM protocols for MNs introduce, especially at high loads. This is achieved by: 1) applying an access algorithm to avoid data wavelengths and receiver collisions, and 2) introducing a multiple bufer architecture at each node with an efective bufer selection for transmission scheme that combines the priority criteria of receiver collisions avoidance and packet age. Thus, we obtain not only dropping probability and delay reduction, but also signifcant throughput enhancement. The required number of bufers per node to maximize throughput is investigated. Performance evaluation is provided by a discrete event simulation model based on self-similar trafc. Comparative results prove that the proposed architecture manages efcient bandwidth utilization as compared with other WDM protocols, especially under high loads. Keword-collsions resoluton; incomng trafc manaement; perormance imrovement; wavelength dvision multipleng. I. INTRODUCTION I literature, the vast majority of Wavelength Division Multiplexing Access (WMA) protocols for ring metropolitan area networks (MANs) achieves limited fber bandwidth utilization due to the phenomena of data chanels ad receiver collisions. Most of tese protocols do not manage to efectively avoid both of them. Especially, in many studies each node uses a fxed tuned transmitter and a tunable receiver and it has a dedicated chanel for transmission to avoid only the channels collisions [1]-[2]. On the conty, in other studies each node uses a tunable transmitter and a fxed tuned receiver and it has a dedicated channel for reception to avoid only the receiver collisions [3]-[4]. Also in these studies, a restriction about the number of nodes around the ring is introduced to provide access faiess: the number of nodes (N) is an integer (D) multiple of the number of chanels (W), D=N I W. This restriction provides serious scalability problems, since in case of a node addition or deactivation the whole network reconfguration is required. I order to improve scalability, the use of a node interface with a fast tunable transmitter and a tunable receiver is introduced in [5]-[6]. Despite the tansceivers tunability benefts, the systems sufer fom high delays. Also, in [1] the scalability is studied ad the substitution of the multiple fxed receivers per node by a tunable one is considered. Although this substitution improves scalability, the system sufers fom high receiver collisions. Some projects like HORET have already experimentally tested the technology restrictions [7]. Improvement in scalability area is achieved in [8], introducing a node interface with a tunable transmitter ad multiple tunable receivers. In this way, independence between the numbers N and W is achieved and the restriction D=N I W is removed. The network uses a separate control channel to efectively exploit the control inforation prior to data packet transmission to avoid both data channels and receiver collisions. Thus, essential performance improvement is achieved as compared to [1], especially in high loads. Despite the scalability and performance enhancement comparatively to other studies, the system utilization of [8] can be improved even more. The key problem is that the system reaches saturation even at low loads providing high packet dropping probability, high queuing delay, ad limited throughput. This is due to te use of a single bufer at each node to organize the incoming trafc. In this way, a node -in order to avoid a receiver collision- cancels the trasmission if the frst packet in the bufer collides at destination with anoter already transmitted packet, although there may exist available data slots for other collision-fee transmissions. I this paper, we extend the study of [8]. Our purose is to improve performance, not only enhancing throughput but also reducing dropping probability and delay experienced. Thus, in each node a multi-bufer architecture is intoduced to substitute the single bufer of [8]. Also, an efective algorithm is presented to properly allocate the incoming trafc into the multiple bufers according to the destination. Moreover, an