Backfill Scheduler Frequency Control for HPC Peak Power Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High Performance Cluster (HPC) systems face challenges in managing peak power usage efficiently due to the variability in processor speeds across nodes, leading to potential communication hindrances and performance slowdowns, especially when nodes are shared among scheduled jobs.
Innovation Solution
A modified backfill scheduler that allows explicit CPU speed parameter setting by site management or user software, reducing CPU frequency of preceding jobs to fill scheduling voids without delaying their completion, thereby reducing peak power usage across thousands of nodes without impacting overall system performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by stationary object
If nodes manage their own power by manipulating frequency and voltage levels, then power consumption is reduced, but system performance and communication uniformity deteriorate due to different processing speeds across nodes
Solution Approach 1:
The patent applies parameter changes by allowing the scheduler to set CPU frequency parameters at the system level rather than allowing each node to independently manage its own frequency. This centralized parameter control enables uniform frequency settings across all nodes, maintaining consistent processing speeds while still achieving power reduction through scheduled frequency adjustments.
2Productivity
If backfill scheduler runs jobs out of order to increase system utilization, then productivity increases, but peak power consumption increases due to more jobs running simultaneously
Solution Approach 1:
The patent applies preliminary action by having the scheduler proactively plan job execution sequences that anticipate power consumption patterns. The scheduler pre-determines which jobs to backfill and at what frequency levels, allowing it to smooth out peak power consumption while still achieving high system utilization through out-of-order job execution.
Solution Approach 2:
The patent applies dynamics by making the CPU frequency adjustable and time-dependent rather than static. The scheduler dynamically adjusts frequency settings for different jobs and time periods, enabling flexible control over power consumption while maintaining productivity through adaptive frequency management during backfill operations.
3Power
If CPU frequency is reduced to fill scheduling voids, then peak power usage is reduced, but job completion time increases
Solution Approach 1:
The patent applies partial action by reducing CPU frequency only during specific scheduling voids rather than continuously. The scheduler selectively applies frequency reduction only when there are idle periods in the job schedule, allowing jobs to run at full speed when resources are fully utilized while reducing power consumption during underutilized periods.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method is disclosed for reducing peak power usage in a large computer system with multiple nodes by identifying jobs which can be scheduled to run at reduced frequency in order to reduce total power usage during certain time periods. The backfill scheduler of the computer system's operating system performs steps providing for selected jobs on selected nodes of the computer system to be run at reduced frequency such that those jobs are partially processed during previously underutilized holes in the computer system schedule in order to reduce overall peak power during a period of processing.