Randomized token buckets: reducing the buffers required in multiplexors

We study the buffer requirements for N identical token bucket streams that are fed to a multiplexor. If the token bucket streams are synchronized the buffer size requirement at the multiplexor is N·B, where B is the size of the token bucket. We introduce a new variant of token bucket schemes that we call randomized token bucket schemes that insert random delays to prevent the streams from becoming (and remaining) synchronized. We show, using a combination of analysis and simulation that the buffer size is reduced to at most 3√NB with fixed small (10^-6) probability of buffer overflow. Because of the randomization, this result does not depend on assumptions about input traffic patterns. While our analysis has been focused on a simple multiplexor, our research is motivated by the possible use of randomized leaky bucket as a traffic shaping mechanism to reduce buffer size and end-to-end delays across multiple hops in a network