EECS 290Q: Project Abstracts

Statistical Multiplexing of Regulated Traffic in an ATM node
Yuan-Chi Chang

Traditional statistical multiplexing analysis relies on accurate statistical models provided by traffic sources. This is not the case, however, in the worst-case framework where traffic is simply regulated by leaky bucket schemes. One advantage of using deterministic traffic regulators is that it relieves sources from specifying their models. Therefore, model accuracy is not an issue and the network can easily verify if sources obey the QoS contracts, which is probably very difficult to do if statistical models are used.

In this project, I reviewed the current research efforts in the direction of applying simple traffic regulators in the statistical multiplexing context. In particular, I focused on the most recent published result and described its novel approach in dealing with the problem of buffer and bandwidth allocations. The paper pointed out the situations when a traffic source is non-statistically multiplexable (non-SM) and further characterized the nonlinearity of the admissible region if SM and non-SM sources are multiplexed. An implication of this nonlinearity is that the nice additive property of effective bandwidth no longer holds.

Since this research field is fairly new, a number of interesting issues remain open. At the end of the paper, I itemized some of them and discussed possible future directions.

UCB Networking / Matt Siler siler@eecs.berkeley.edu/ April 19, 1996