Cell loss probability for the prioritized leaky bucket

The paper studies the prioritized leaky bucket (PLB) as a usage/network parameter control (UPC/NPC) mechanism. The UPC/NPC mechanism is an important part of the overall congestion control strategy for B-ISDN using asynchronous transfer mode (ATM). The PLB is a modified leaky bucket mechanism which incorporates selective cell discarding to manage prioritized traffic for an ATM connection. Selective cell discarding is accomplished by using a prioritized token pool or a prioritized input buffer. The PLB is analyzed as a single server with a prioritized queue using the fluid flow approximation method. The source is modeled by an on/off fluid-flow source. The burst and silence periods are both exponentially distributed. By forming a Markov chain of the buffer state at the beginning of each burst, a set of integral equations can be obtained. Simulations are then taken to examine the validity of the fluid-flow analysis. Three types of prioritized traffic source models are considered. They are the non-biased, small biased, and large biased source models. We conclude that the accuracy of the fluid flow analysis decreases with the amount of bias introduced to the source model. The cell loss probabilities derived from the fluid flow analysis are valid for the non-biased and small biased cases. However, they fail to accurately reflect the cell loss in the large biased case. The simulation results also show that the small biases source benefits more from the PLB. This observation provides a new insight on the design of a priority coding scheme for an ATM connection