A Data Driven Scheduling Approach for Power Management on HPC Systems
SessionMemory and Power
Session ChairRolf Riesen
Event Type
Paper
Intermediate
Power
System Administration
System Software
TimeWednesday, November 16th4:30pm - 5pm
Location355-BC
DescriptionSchedulers running on HPC systems traditionally consider the number of resources and the time requested for each job for scheduling. However as systems get larger, other metrics like power become necessary to ensure system stability.
In this paper, we propose a data driven scheduling approach for controlling the power consumption of the entire system under any user defined budget. Our approach actively observes, analyzes, and assesses power behaviors of the system and user jobs to guide scheduling decisions for power management. This design is based on the key observation that HPC jobs have distinct power profiles. This work contains an empirical analysis of workload power characteristics on a production system, dynamic learner to estimate job power profiles for scheduling, and an online power-aware scheduler for managing overall system power. Using real workload traces, we demonstrate that our design effectively controls system power consumption while minimizing the impact on system utilization.
In this paper, we propose a data driven scheduling approach for controlling the power consumption of the entire system under any user defined budget. Our approach actively observes, analyzes, and assesses power behaviors of the system and user jobs to guide scheduling decisions for power management. This design is based on the key observation that HPC jobs have distinct power profiles. This work contains an empirical analysis of workload power characteristics on a production system, dynamic learner to estimate job power profiles for scheduling, and an online power-aware scheduler for managing overall system power. Using real workload traces, we demonstrate that our design effectively controls system power consumption while minimizing the impact on system utilization.
Download PDF
Paper provided by the IEEE Computer SocietyPaper also available from the ACM Digital Library
Authors
