A Study On Adaptive Real-Time Scheduling

An increasing number of real-time applications are more and more complex and flexible. The characteristic of these applications related to multimedia, adaptive control and videoconference and so on is dynamic and time-varied. So, the emphasis of research on real-time system is from static scheduling to dynamic scheduling, from predictable environment to unpredictable environment, from periodic tasks to aperiodic tasks. This thesis gives emphasis to the real-time of dynamic system, and the main contents are as follows.Firstly, it gives a brief review to the research status of dynamic system real-time scheduling; mostly introduces the feedback control real-time scheduling and real-time queueing.Then, it proposes a static resource management method. This method can generate the upper and the lower bounds of tasks in linear time complexity in the absence of rate constraints, and find a feasible solution of task set. And then we design a very simple stochastic heuristical algorithm to maximize the weighted reward of task set. The better approach is to compute an initial solution by our stochastic heuristical algorithm and then search the more precise solution by other algorithms such as Genetic Algorithm, Simulating Anneal, and Artificial Neural Network and so on. The feasible solution can also be got by other methods.And then, it discusses some results on feedback real-time scheduling. The self-regulation in the feedback control theory is first introduced into the dynamic real-time system. A dynamic real-time scheduling model based on self-regulation used to single processor aperiodic task set is proposed. This model adopts a novel method modeling real-time system to integrate admission control with QoS degradation, and uses feedback control technique to meet timeliness requirements, and improve the system throughput. And then, proposes a linear quadratic stochastic optimal control model used to the end-to-end distributed real-time system. This model uses two feedback loops to guarantee the system's performance.And then, it discusses a method based on real-time queueing. The aperiodic soft real-time system where tasks' arriver time and execution time are stochastic can be regarded as a stochastic service system. Firstly, we construct a dynamic optimal control model for FCFS scheduling strategy; then, we present the more useful model for DM scheduling algorithm, so, the aperiodic soft real-time system is a single server queue system with multiple priorities. The model can statistically guarantee the schedulability of real-time systems, at the same time it can guarantee the QoS(Quality of Service) and improve throughput of systems. Finally, analyses the influence of buffer threshold on system's performance, and designs the control model with buffer threshold.Lastly, it introduces an instance research, namely, the end-to-end transmission of video stream from wired channel to wireless channel. The target is to guarantee the real-time transmission, at the same time guarantee the playout quality. We propose a frame layer adaptive rate control algorithm for the wireless transcoder based on the TMN8 test model. The algorithm adopts stochastic control technique to adapt the variation of wireless bandwidth. We analyze the performance in detail. Experiment results demonstrate that the algorithm can improve the synthetic performance of rate control through the compromise between the end-to-end delay and the playout quality. To be compared with TMN8, the proposed algorithm can guarantee the continuity of video communication under the condition of great variation of bandwidth.