ABSTRACT NCR-9730105 Hui Zhang Carnegie Mellon University Rick Wilder and Chuck Song MCI Joint with NCR-9730104 Edward W. Knightly Rice University Joint with NCR-9730103 Jorg Liebeherr Polytechnic University Integrated Resource Management for the Integrated Services Internet The IETF has defined several key components for enabling an integrated services Internet: (1) service classes, which define the service models for supporting end-to-end per-flow QoS, (2) a link sharing model, which defines how different organizational entities (agencies) can share the same physical resources, and (3) a resource reservation protocol, which provides the signaling mechanisms to support these services. While these capabilities are indeed revolutionary as compared to the current Internet, serious impediments remain to achieving the vision of a ubiquitous multi-service Internet. In particular, efficient resource management algorithms must be available that actually deliver these services to the applications. Packet scheduling algorithms must be devised that can meet the service requirements of heterogeneous applications and different agencies while, at the same time, exploiting statistical multiplexing; admission control algorithms must be designed to allocate the minimum amount of network resources without violating any of the service requirements. At present, no set of resource management algorithms is available that can provide multi-service capabilities while considering the complex dynamics among connections, services, and agencies. This project will provide the first truly integrated resource management infrastructure for multi-service networks. Taking a new and comprehensive approach at resource management algorithms within the framework of the integrated services Internet, we propose to develop: (1) A novel hierarchical link sharing service model and algorithm that simultaneously supports all of the services proposed by the IETF without sacrificing efficient utilization of network resources, (2) Traffic control and scheduling algorithms that are amenable to efficient implementation in high speed and scalable switching systems, (3) Admission control algorithms for an enforceable controlled-load service with statistical quality of service guarantees, (4) Robust admission control algorithms for measurement-based controlled-load services, and (5) Traffic control algorithms that provide graceful quality-of-service adaptation in the presence of transient fluctuations of the network load. This project will develop the theoretical underpinnings of integrated resource management algorithms for multi-service networks. Moreover, in a collaboration that involves an integrated services network provider (MCI Telecommunications), a network equipment vendor (Ascend), and academic researchers (CMU, Rice, Polytechnic), the PIs will build a complete prototype implementation of the proposed algorithms on the vBNS national network infrastructure. This reference implementation will provide a first-of-its-kind platform for experimental research on the integrated services Internet that will yield critical input to both practitioners in the field and standards bodies on the effectiveness of current and newly proposed resource management algorithms.
