Skip to content Skip to main navigation Report an accessibility issue

Optimizing System Scheduling Parameters in Heterogeneous Computing Environments

spowers_headshot1

Dr. Sarah Powers
Research Staff Member
Computer Science and Mathematics Division
Oak Ridge National Laboratory
September 18, 2015, 2:30 – 3:30 PM
410 John D. Tickle Engineering Building

Dr. Sarah Powers is a research staff member at Oak Ridge National Laboratory in the Computer Science and Mathematics Division. Working at the inter- section of mathematics, computer science and statistics, her research interests span a variety of areas related to Operations Research (OR) and applied mathematics including optimization, algorithm development and scalability, data analytics, machine learning and simulation. She enjoys using mathematics to find more effective ways to achieve results, make decisions and enabling others to understand the tools at their disposal to do the same. She holds a M.S. in Operations Research and Statistics and Ph.D. in Decision Sciences and Engineering Sciences from Rensselaer Polytechnic Institute and a B.S. in Mathematics from Gordon College.

Abstract: A wide variety of computational resources exist today, varying from super- computers such as Titan at Oak Ridge National Laboratory (ORNL) to grid computing clusters spread across geographical locations. As High Performance Computing (HPC) continues to expand into the Exascale realm, the computational capability, heterogeneity and workload diversity of these systems increase. Optimizing resource usage is key while meeting user demands and minimizing cost (e.g., energy consumption). This talk will discuss a tool for exploring system-level scheduler parameter settings in a heterogeneous computing environment. Through the coupling of simulation and optimization techniques, this work investigates optimal scheduling intervals, the impact of job arrival prediction on scheduling, as well as how to best apply fair use policies. The developed simulation framework is quick and modular, enabling decision makers to further explore decisions in real-time regarding scheduling policies or parameter changes.