Adaptive QoS and resource management using a posteriori workload characterizations

Certain real-time applications must operate in highly dynamic environments (e.g., battle environments), thereby precluding accurate characterization of the applications´ workloads by static models. Thus, static guarantees of real-time performance based on a priori characterizations are not possible. However, potential benefits of a posteriori approaches are significant, including the ability to function correctly in dynamic environments (through adaptability to unforeseen conditions), and higher actual utilization of computing resources. In this paper, we present an approach that is appropriate for systems which experience large variations in workload. A distributed collection of computing resources is managed by continuously computing and assessing QoS and resource utilization metrics that are determined a posteriori. The utility of our approach is shown by applying it to a large, experimental distributed Navy computing system