2016-04-12 SIAM'16 (16th Conference on Parallel Processing for Scientific Computing) Invited Talk: The Influence of Data Center Power and Energy Constraints on Future HPC Scheduling Systems

Overview

Reducing the power and energy consumption of HPC data centers is one of the essential areas of exascale research. Currently, PC data centers are built and provisioned for expected peak power. But the standard workload power consumption can be as low as 50% of peak power. Therefore, future HPC data centers would then over-provision or right size their data center; meaning that power and cooling is only sufficient for normal operation. Additional power restrictions might come from the power providers (maximum power change in a specific time frame). Other factors could include, for example, peak power vs. energy consumption over the year, real time power consumption adjustments based on renewable energy generation, and moving from CPU hour based accounting to energy consumption. Since scheduling systems are currently the main way to allocate resources for applications, it seems clear that power and energy might need to become a scheduling resource/parameter. This talk will discuss possible data center policies concerning power and energy and their impact on scheduling decisions.