Adaptive QoS Optimizations with applications to Radar Tracking ⋆ Sourav Ghosh 1 , Jeffery Hansen 2 , and Ragunathan (Raj) Rajkumar 3 and John Lehoczky 4 1 Carnegie Mellon University, Department of Electrical and Computer Engineering sourav@cs.cmu.edu 2 Carnegie Mellon University, Institute for Complex Engineered Systems hansen@cmu.edu 3 Carnegie Mellon University, Department of Electrical and Computer Engineering raj@ece.cmu.edu 4 Carnegie Mellon University, Department of Statistics jpl@stat.cmu.edu Abstract. In many applications such as sensor networks, mobile ad hoc net- working and autonomous systems, the relationship between level of service and resource requirements is not fixed. Environmental factors outside the direct con- trol of the system affect this relationship and may also affect the perceived utility of a given level of service. Radar tracking provides a good example. In radar systems, a fixed amount of radar bandwidth and computing resources must be apportioned among multiple tasks, each of which corresponds to a target. In ad- dition, environmental factors such as noise, heating constraints of the radar and the speed, distance and maneuverability of tracked targets dynamically affect the mapping between the level of service and resource requirements as well as the mapping between the level of service and the user-perceived utility. To be able to handle these tasks, a QoS manager must be adaptive, reacting to dynamic changes in the environment, adjusting the level of service and reallocating resources effi- ciently. In this paper, we present a dynamic QoS optimization scheme for a radar tracking application based on Q-RAM [1, 2]. Our scheme is able to deal with a large number of operating points in real-time with very acceptable losses in total utility accrued. This result is made possible by an efficient heuristic to compute the concave majorant of a multi-variate function, and an off-line storage-efficient discretization of the static aspects of the problem space. These two contributions will also be useful in many dynamic QoS-driven applications beyond radar track- ing. 1 Introduction Traditional QoS optimization algorithms assume that a collection of tasks compete for resources, with each task receiving some benefit from those resources. The goal is to allocate the resources to tasks in such a way as to optimize the total benefit received by all the tasks. This benefit is often called “utility” [3]. A larger QoS for a task generally ⋆ This work was supported by a DARPA Multidisciplinary University Research Initiative (MURI) program administered by the Office of Naval Research under Grant N00014-01-1- 0576 and by DARPA under contract number F33615-00-C-1729.