Discrete time analysis of leaky-bucket congestion control

The authors investigate the effectiveness of the leaky bucket as a traffic smoother and the effect on the system performance as the system complexity, i.e., the size of the data buffer and token pool, increases. The goodness of a controller is assessed by performance measures such as packet loss rate, queueing delay, and burstiness of the departure process. Using a discrete-time model, recursive algorithms for evaluating those performance measures of a leaky bucket controller are developed. It is shown that a small-to-medium-sized leaky bucket is sufficient to attain the control capability. Also, it is shown that at traffic loads where the data loss probability is not significant, the leaky bucket is not so effective as a traffic smoothing device in that the ratio of the variance of interarrival times to that of the interdeparture times is not much less than unity