Scheduling Bag-of-Task Jobs with Selective Approximate Computations in Distributed Systems

In distributed systems, a wide range of parallel applications have a bag-of-tasks (BoT) structure with various characteristics. Due to their computational demands and time constraints, their efficient scheduling is becoming more challenging. We propose two approximate computations-based scheduling approaches so that we address this issue. Both of them incorporate imprecise computations in a selective manner. Through experimentation, we evaluate the system performance under the considered scheduling schemes, utilizing different load levels and BoTs characteristics. The experimental findings indicate that the implementation of selective imprecise computations yields a notable enhancement in system performance while exhibiting only a negligible decrease in result precision. Additionally, the results offer insights into the performance of the analyzed scheduling methods across each of the investigated scenarios.