Node Power Capping in HPC Clusters Based on Job Type
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing power capping mechanisms in high-performance computers (HPCs) cause significant performance degradation by uniformly applying power limits across nodes, which is inefficient for jobs with varying resource demands, particularly memory-bound, compute-bound, and mixed-type jobs.
Innovation Solution
A method that assigns power caps based on the type of job (memory, compute, or mixed) and adjusts existing caps to balance overall HPC power consumption while minimizing performance impact, using predefined values and adaptive capping techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by stationary object
If uniform power capping is applied to all nodes, then overall power consumption is limited, but job performance is significantly degraded
Solution Approach 1:
The patent applies different power cap values to different nodes based on their specific job types. Memory-bound jobs receive higher power caps while compute-bound jobs receive lower power caps, creating localized quality differences in power allocation that optimize both energy efficiency and job performance
Solution Approach 2:
The system dynamically changes the power cap parameter based on job characteristics. By identifying job types (memory-bound, compute-bound, mixed) and assigning appropriate power cap values, the system adapts power consumption parameters to match actual workload requirements, resolving the contradiction between power limitation and performance maintenance
2Use of energy by stationary object
If power cap values are reduced to meet power budget, then power consumption compliance is improved, but computing power is reduced
Solution Approach 1:
Different nodes receive different power cap values tailored to their job types. Memory-bound nodes can operate at higher power levels while compute-bound nodes operate at lower power levels, ensuring power budget compliance while maintaining optimal computing power distribution across the system
Solution Approach 2:
The power cap values are dynamically adjusted based on the type of job being executed on each node. The system transitions from static uniform power capping to dynamic adaptive power capping, where power allocation changes in response to workload characteristics, maintaining both power compliance and computing effectiveness
Data Source
Figure 1~2
AI summary
According to an aspect of the invention, it is provided a computer implemented method for capping the power consumption of a high-performance computer comprising a plurality of nodes, the plurality of nodes comprising a first group of nodes of which each node is allocated to a first job, the power capping of each node of the first group of nodes being enforced to a first capping value being dependent on the type of the first job.