ATM and next generation IP networks will provide the fundamental building blocks for integrated services networks: the ability to reserve network resources on behalf of bursty real-time applications. However, provisioning network resources accurately is quite challenging in the real-world environment characterized by heterogeneous and highly bursty traffic streams which exhibit rate variations over multiple time scales. The goal of this research is to design accurate and robust admission control and resource management algorithms that scale to large internetworks. In particular, we plan to study the following three problems. First, in a shared public network, applications cannot be trusted to adhere to their specified traffic parameters and therefore must be policed. However, deterministic policing must not preclude statistical resource sharing, which is essential for efficient resource utilization. We are designing an accurate envelope-based resource allocation algorithm that enables the network to statistically multiplex traffic streams while simultaneously enforcing applications' specified traffic parameters. Second, many applications will be unable to specify their traffic parameters in advance so that the network must somehow reserve resources even for applications with unknown resource requirements. We are developing a measurement-based service that adaptively measures the envelope of the aggregate traffic flow to predict future resource requirements and performance parameters for such ill-characterized applications. Finally, internetworks are characterized by a large number of heterogeneous components so that a resource management scheme must scale to such realistic environments. We are devising end-to-end resource management techniques that exploit coarse-grained representation and allocation of network resources.
