The IEEE 802.5 token ring allows the user to define multiple message priorities. The priority mechanism is implemented using a reservation scheme whereby packets attempt to reserve the token. We present an analytic model that predicts mean delay for each priority level. We show that operation of the priority mechanism is not free; that is, there is a well-defined cost (measured in terms of increased average message delay) associated with its implementation, and that in general there is minimal discrimination made between the various priority levels. We also show that the cost of the priority mechanism is an easily computable function of various network parameters. One of the circumstances where the priority mechanism results in a large discrimination between priorities is when service is limited to a single packet per received token. We show that single-packet-per-token service is necessary (but not sufficient) to gain maximum benefit from priority operation. Our main conclusion is that the network user will not, in general, receive much benefit from the priority mechanism since the priority mechanism can in fact <italic>increase</italic> slightly the delay for all priorities. The exception to this statement occurs when the network is operating under fairly extreme conditions which are well defined, and which we believe are unlikely in most installations.
Unfortunately, ACM prohibits us from displaying non-influential references for this paper.
To see the full reference list, please visit http://dl.acm.org/citation.cfm?id=55490.