Calculating Task Output Event Models to Reduce Distributed System Cost

Input event models such as periodic, jitter, and sporadic are known as powerful abstractions for task workload arrival in scheduling analysis. In distributed systems, such task input models result from output event models of other tasks in the system. Interestingly, output event models have received only little attention in literature. In this paper, we introduce a novel approach to determine task output event models for the most practically important task configurations. We show that the output event model accuracy is key to efficient system dimensioning with a direct impact on system cost, and we propose several optimizations. The experiments show that the new approach reveals previously unknown, and hence, unexploited system performance reserves.