Research Assistant: K. He
Sponsor: NSF
With advances in multimedia and networking technologies, applications on general purpose computers are getting more diverse. Two important characteristics distinguish emerging applications, which typically employ forms of continuous media, from more traditional ones. First, many of them are increasingly dependent on access to heterogeneous resources distributed over a local-area or wide-area network. Second, these applications often have well defined, though adaptive, quality of service (QoS) constraints. Building on our existing results in single resource (i.e. CPU and bandwidth) schedulers and operating system abstractions for predictable performance, we are investigating a unified QoS architecture for emerging general purpose network computing. We investigate resource management issues interesting in their own right: in CPU scheduling, we have designed and built adaptive fair share and deadline based schedulers, and are studying their simultaneous application for decoupled delay and rate guarantees; in protocol processing, we are investigating interprocess communication (IPC) abstractions with QoS properties, and protocol implementation techniques that minimize hidden scheduling in network access. In addition, we aim to synthesize related research results into an innovative computing platform called Concerto. Concerto extends traditional network level resource management, such as end-to-end signaling and route optimization, to resource types other than network bandwidth and connectivity. Contents profiles and QoS brokers in Concerto enable domain specific knowledge to be exploited in adapting resource requirements to resource availability. Further productive adaptations are made possible with performance monitors and application knobs integrated with QoS aware applications. These research activities are being conducted in the Systems Software and Architecture Laboratory.