Performance Evaluation of Non-Preemptive Earliest Deadline First Scheduling Techniques

The deadlines associated with periodic tasks are of great importance to the functionality of a real-time system. To fulfill the deadline requirement of real-time systems, various scheduling policies are applied. Among these scheduling algorithms, the earliest deadline first scheduling algorithm is an optimal technique under preemptive class which promises 100% system utilization. Unfortunately, this approach does not exhibit optimal performance when considered for non-preemptive systems. Due to its wider acceptance in industry, the earliest deadline first scheduling algorithm has been studied extensively as an optimal scheduling algorithm and several feasibility techniques do exist to validate the timing constraints of the system. In this work, we extend our previous results and study the behavior of closely related feasibility tests under varying system utilizations for the non-preemptive earliest deadline first algorithm and show that the performance of a non-preemptive scheduling algorithm degrades significantly at a system utilization of 80% or higher, irrespective of the scheduling policy.