A Distributed System Reference Architecture for Adaptive QoS and Resource Management

1 The Naval Surface Warfare Center, Dahlgren, VA, 22448, {MastersMW|WermePV|MaddenLA|MarlowDT|IreyPM}@nswc.navy.mil. 2 Computer Science and Engineering Dept., The University of Texas at Arlington, Arlington, TX 76019-0015, {welch|shirazi}@cse.uta.edu. Sponsored in part by DARPA/NCCOSC contract N66001-97-C-8250, and by the NSWC/NCEE contracts NCEE/A303/41E-96 and NCEE/A303/50A-98. This paper deals with large, distributed real-time systems that have execution times and resource utilizations which cannot be characterized a priori. (The motivation for our work is provided in part by the characteristics of combat systems, as described in [1].) There are several implications of these characteristics: 1) demand space workload characterizations may need to be determined a posteriori, and 2) an adaptive approach to resource allocation may be necessary to accommodate dynamic workload changes. Thus, we present a scheme for dynamically managing distributed computing resources by continuously computing and assessing QoS and resource utilization metrics that are determined a posteriori. Specifically, the paper presents an adaptive distributed system reference architecture that is suitable for such an approach. This reference architecture provides the capabilities and infrastructure needed to construct multi-component, replicated, distributed object real-time systems that negotiate for a given level of QoS from the underlying distributed computing resources.