Performance Model for HPC Parameter Optimization Under Power Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High Performance Computing (HPC) systems face increased power consumption as performance scales up, leading to power restrictions that require users to adjust parameter values of application programs to achieve optimal performance, but these values are not universally effective across different power constraints.
Innovation Solution
An information processing apparatus generates a performance model based on data from executing application programs under various power restrictions, allowing it to calculate and output optimal parameter settings for achieving the highest performance value under specific power limitations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the performance of HPC system is enhanced, then the computation capability is improved, but the consumption power increases
Solution Approach 1:
The patent changes parameters of the application program (such as loop unrolling factors, block sizes, parallelization degrees) to optimize performance under different power restrictions. By adjusting these parameters, the system achieves different performance levels corresponding to different power consumption levels, resolving the contradiction between computation capability and power consumption.
2Productivity
If parameter values of application program are adjusted to achieve sufficient performance under limited power, then the performance under power restriction is improved, but the same parameter values are not effective under different power restrictions
Solution Approach 1:
The patent introduces a dynamic parameter adjustment mechanism that automatically changes application program parameters based on the detected power restriction level. The system monitors the current power restriction and dynamically selects or adjusts parameters to optimize performance for that specific condition, making the system adaptable to varying power restrictions rather than using fixed parameters.
Solution Approach 2:
The system implements a feedback mechanism where the actual power consumption and performance metrics are monitored, and this information is used to adjust parameters for subsequent executions. The patent uses measured performance data under different power restrictions to determine optimal parameter settings, creating a closed-loop control system that adapts to power conditions.
3Productivity
If manual adjustment of parameter values is performed to optimize performance, then the performance under specific power restriction is improved, but the complexity of operation increases
Solution Approach 1:
The patent implements a self-service mechanism where the system automatically performs parameter optimization without requiring manual user intervention. The system autonomously measures performance under different power restrictions, analyzes the data, and determines optimal parameter settings, eliminating the need for users to manually adjust complex parameters while still achieving optimized performance.
Data Source
AI summary
An information processing apparatus includes a memory; and a processor coupled to the memory and configured to generate a performance model for calculating a performance value of an application program from a power restriction for each set of parameters of the application program, based on data acquired when a computing apparatus executes the application program for each set of parameters of the application program under each of a plurality of power restrictions; calculate, for each set of parameters of the application program, the performance value of the application program from a first power restriction different from any of the plurality of power restrictions, based on the performance model generated for each set of parameters of the application program; and output a set of parameters of the application program corresponding to a highest performance value of the calculated performance values.