Optimal state prediction for feedback-based QoS adaptations

In heterogeneous network environments with performance variations present, complex distributed applications, such as distributed visual tracking applications, are desired to adapt themselves and to adjust their resource demands dynamically, in response to fluctuations in either end system or network resources. By such adaptations, they are able to preserve the user-perceptible critical QoS parameters, and trade off non-critical ones. However, correct decisions on adaptation timing and scale, such as determining data rate transmitted from the server to clients in an application, depend on accurate observations of system states, such as quantities of data in transit or arrived at the destination. Significant end-to-end delay may obstruct the desired accurate observation. We present an optimal state prediction approach to estimate current states based on available state observations. Once accurate predictions are made, the applications can be adjusted dynamically based on a control-theoretical model. Finally, we show the effectiveness of our approach with experimental results in a client-server based visual tracking application, where application control and state estimations are accomplished by middleware components.