A Scheduling Algorithm for Many-Task Computing Optimized for IO Contention in Heterogeneous Grid Environment

Many-task computing (MTC) is a computing paradigm widely used in scientific area. Each MTC job consists of up to millions of independent tasks. The primary strategy of improving MTC efficiency is to execute different tasks on parallel computing resources. Some MTC applications are data-dependent. For such an application, some specified data resource is required by every task; hence access to the data resource is its performance bottleneck. To mitigate IO contention among tasks executed in parallel, this paper presents a scheduling algorithm. The experimental results show that our algorithm achieves better performance in both makespan and resource utilization when compared with other works.