Dynamic CPU Scheduling with Imprecise Knowledge of Computation-Time

The majority of the studies conducted in scheduling real-time transactions mostly concentrate on concurrency control protocols, while overlooking the CPU as being the primary resource. Consequently, there are various techniques for scheduling the CPU in conventional time-critical systems; meanwhile, there does not seem to be any technique that is adequately designed for scheduling such a resource in Real-Time Database (RTDB) systems. Existing schemes for scheduling the CPU in RTDB systems considerably degrade under overload conditions, in addition to suffering a relatively high preemption rate under all operating loads. Coping with overload situations requires accurate knowledge of the computation time, and preemption must be augmented with synchronization protocols, which contribute to performance degradation as data contention rises – aside from context switching overhead. In this study, we construct an efficient CPU scheduling scheme that minimizes the preemption rate in order to reduce the frequency by which synchronization protocols must be invoked, along with their inherited performance degradation. In addition, we also introduce a new timing model upon which the newly introduced scheduler is incorporated in order to utilize the system’s imprecise knowledge of computation time estimates.