Smart Job Scheduling for High-Performance Cloud Computing Services

In this paper, we describe the challenges faced and lessons learned while establishing a large-scale high performance cloud computing service that enables online mechanical structural analysis and many other scientific applications using the finite element analysis (FEA) technique. The service is intended to process many independent and loosely-dependent (e.g. assembled system) tasks concurrently. Challenges faced include accurate job characterization, handling of many-task mixed jobs, sensitivity of task execution to multi-threading parameters, effective multi-core scheduling in a single node, and achieving seamless scale across multiple nodes. We find that significant performance gains in terms of both job completion latency and throughput are possible via dynamic or "smart" partitioning and resource-aware scheduling compared to shortest first and aggressive job scheduling techniques. We also discuss issues related to secure and private processing of sensitive models in the cloud.