Adaptive Fuzzy Control for Utilization Management

An increasing number of real-time systems are embedded in mission critical systems such as target tracking systems, in which workloads may dynamically vary, for example, depending on the number of targets in the area of interest Feedback control has been applied to support real-time performance in dynamic environments, producing promising initial results. However, mathematical system modeling necessary for feedback control is challenging. To reduce the difficulty of system modeling, we apply fuzzy control for direct nonlinear mappings between the utilization error (= target utilization - current utilization) and the workload adjustment required to achieve the target utilization via IF-THEN rules. Moreover, via online adaptation, our fuzzy controller can amplify or dampen its own fuzzy control signal, if necessary, to expedite the convergence to the desired utilization. In our simulation study, our approach quickly converges to the target utilization when the workload significantly changes. In contrast, the tested baselines oscillate between overload and underutilization.