Supervisor: Zhiling Lan (Illinois Institute of Technology)
Abstract: High-performance computing (HPC) is undergoing significant changes. Next generation HPC systems are equipped with diverse global/local resources. HPC job scheduler plays a crucial role in efficient use of resources. However, traditional job schedulers are single-objective and fail to efficient use of other resources. In our previous work, we present a job scheduling framework named BBSched to schedule CPUs and burst buffers. As we are heading toward exascale computing, a variety of heterogeneous resources are deployed in HPC systems. In this poster, we extend BBSched for managing multiple resources beyond CPUs and burst buffers. We formulate multi-resource scheduling as a general multi-objective optimization (MOO) problem, present a heuristic method to solve the NP-hard MOO problem, and provide a preliminary evaluation for scheduling up to ten resources. The proposed multi-resource scheduling design is intended to enhance Cobalt, a production job scheduler deployed on HPC systems at Argonne Leadership Computing Facility (ALCF).
ACM-SRC Semi-Finalist: no
Poster Summary: PDF
Back to Poster Archive Listing