Executing Dynamic and Heterogeneous Workloads on Super Computers
André Merzky, Mark Santcroos, Matteo Turilli, Shantenu Jha · arXiv (Cornell University) · 2015
Many scientific applications have workloads comprised of multiple heterogeneous tasks that are not known in advance and may vary in the resources needed during execution. However, high-performance computing systems are designed to support applications comprised of mostly monolithic, single-job workloads. Pilot systems decouple workload specification, resource selection, and task execution via job placeholders and late-binding. Pilot systems help to satisfy the resource requirements of workloads comprised of multiple tasks with the capabilities and usage policies of HPC systems. RADICAL-Pilot (RP) is a portable, modular and extensible Python-based Pilot system. In this paper we describe RP’s design, discuss how it is engineered, characterize its performance and show its ability to execute heterogeneous and dynamic workloads on a range of high-performance computing systems. RP is capable of spawning more than 100 tasks/second and the steady-state execution of up to 8,000 concurrent tasks. RP can be used stand-alone, as well as integrated with other application-level tools as a runtime system.