Using feedback control for a network and CPU resource management application

Interactive multimedia and sensor-based real-time applications on the Internet must be delivered to the receiver at the sender's transmission rate and have bounded end-to-end delay requirements. We call such flows real-rate flows. The resource requirements of such flows can not be met in many environments. Researchers have proposed selectively dropping data in the flow so that the data rate matches available resource capacity (i.e., QoS adaptive, real-rate flows). Application-level QoS adaptation is not well integrated with network scheduling mechanisms and resource management, resulting in slow and inaccurate QoS adaptation, poor flow quality, and unpredictable interactions between adaptive mechanisms. Our proposed resource management service integrates QoS adaptation and network/CPU scheduling mechanisms to offer support for QoS adaptive, real-rate flows. Its goal is to deliver the highest quality flow to the receiver within a bounded delay at the sender's real rate. This service combines two techniques - both based on feedback real-rate scheduling and QoS adaptation. Real-rate scheduling uses packet time-stamps to schedule flows along the flow path at the sender's real rate. QoS adaptation drops data at a node when the node's resources are limited, in application-specified priority order. System modeling issues are discussed and a controller structure is proposed.