Adaptive fair resource management with an arbiter for multi-tier computing systems
ETFA'09 Proceedings of the 14th IEEE international conference on Emerging technologies & factory automation
| Hi-index | 0.43 |
A window based on-line scheduling algorithm to efficiently distribute processor service time to tasks for processing image frames is proposed for the Internet video conferencing system. The algorithm lets the tasks share a limited service time in a manner that the overall quality of the conferencing system is maximized. A task for processing an image frame is modeled as an IRIS (increasing reward with increasing services) task and its reward is mapped into a strictly concave function. Based on the modeling and an observation that the tasks have a strong dependency in their execution order, the scheduler schedules only tasks in a small window at each scheduling instant. And the small window in which a few tasks are contained lets the algorithm have a very low computational complexity, which is a very important consideration for time constraint applications like the Internet conferencing system. Even with low complexity, the scheduler nearly maximizes the total accrued reward as obtained by an on-line optimal algorithm and enhances smooth display, which may be another important quality measure of the conferencing system. To evaluate the quality, three metrics are suggested and the performance of the proposed algorithm is measured in aspects of the metrics.