A Software Licenses Aware Job Scheduling and Management Approach on Multi-clusters

摘要

For commercial software in scientific and engineering computing, software licenses are needed when running them in high performance computing systems. Usually, there is a constraint for the number of software licenses. With the traditional software licenses management approaches, there is a prominent issue. The jobs will fail immediately without available software licenses. However, the existing job scheduling mechanisms mainly focus on hardware resources allocation. In this paper, we propose a software licenses aware job scheduling and management approach on multi-clusters, which is a common scenario for current high performance computing community. Based on floating licenses and scheduling policies, the submitted jobs from multi-clusters requiring software licenses will be uniformly queued and scheduled. A co-scheduling mechanism is provided for the software licenses and hardware resources. So software and limited licenses can be shared and used efficiently on multiclusters. We implemented and evaluated the approach in a real multi-clusters computing environment.