ABSTRACT
The feedback-based dynamic resource allocation heuristics impose some requirements on the target system. Usually, to perform resource allocation decision one can rely on various metrics provided by the monitoring infrastructure tools and services, such as utilization and the time latency between input and output timestamps. To tune the parameters of the controller, the task slack growth response on a step-input in the open-loop system (i.e., without any feedback) has been analysed. This is a typical way in control-theory-based approaches. The obtained results have confirmed the accumulating (integrating) nature of the process, and thus the accumulating process version of Approximate M-constrained Integral Gain Optimisation tuning formulas have been applied to choose the proper values of proportional-integral-derivative controller components. In dynamic workloads, admitting and executing all hard real-time tasks belonging to a taskset can jeopardise system timeliness.
