Scheduling Different Types of Gang Jobs in Distributed Systems
Georgios L. Stavrinides, Helen D. Karatza · 2019
Efficient resource allocation and job scheduling are key factors when evaluating the performance of large-scale distributed platforms. The case of scheduling parallel jobs of gang type, along with regular local jobs, is especially difficult in such systems. In this research, we study algorithms for the scheduling of gang jobs in a cluster of distributed processors, where local jobs have higher priority for execution and they do not negotiate the resources with gangs. We examine different patterns of gang jobs degree of parallelism and their impact on the performance of gang scheduling policies. We use simulation in order to evaluate the performance under different workload variability cases. Our simulation experiments reveal that the performance of the examined gang scheduling algorithms is dependent on the characteristics of both types of jobs in the system, gangs and local jobs.