A multiclass preemptive protocol for high speed local and metropolitan area networks

Future high speed local area networks (LAN) and metropolitan area networks (MAN) will support different classes of traffic (data, voice, video, etc.), each with different quality of service requirements. Most of the proposed protocols supporting multiclass traffic work on the basis of bandwidth reservation. Packet-by-packet based reservation increases the transmission latency of a packet and decreases the throughput of the network, whereas, static allocation fails to respond to the dynamic changes in the bandwidth requirements. We consider a slotted ring based LAN/MAN handling multiple classes of traffic with different priorities. We propose a media access control protocol that gives preemptive priority to higher classes of traffic and requires no bandwidth reservation. It guarantees bounded delay delivery of packets belonging to the highest priority class and ensures statistical bounds on delays for other classes. The protocol is distributed in the sense that any station can locally determine if it can use a slot, without exchanging any message with other stations. It is adaptive to the load distribution of the system. We discuss an efficient implementation of the protocol. The protocol is studied for multiple classes of traffic, and in particular, for voice-data integration in a LAN/MAN environment. Queueing theoretic and matrix analytic techniques are used to derive closed form expressions for the approximate distributions of the response times of different classes of traffic. This also provides other performance metrics of interest (statistical bound on jitter, probability of clipping, etc.). Simulations are conducted to confirm the analytical results. A comparative study with other proposed ring protocols (FDDI, Cambridge Fast Ring, TORNET2) is performed. The study shows that the preemptive protocol exhibits better performance over a wide range of workloads. The slot preemption technique is applied to hard real-time message communication. The non-real-time traffic is capable of using any bandwidth unused by the real-time traffic without affecting the schedulability of the real-time traffic. We propose a distributed implementation of the slot allocation scheme and derive a simple schedulability condition for the same.