Job Power Estimation Using Segmented Recurrent Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high-performance computing systems, accurately estimating power consumption is challenging due to varying execution times of jobs and diverse patterns of power consumption, leading to low accuracy in existing estimation models.
Innovation Solution
The method involves generating multiple estimation models for different time periods based on historical power consumption data, using a recurrent neural network to improve the accuracy of power estimation by reducing variations in data sets and converging learning results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single estimation model is used for all time periods, then the device complexity is reduced, but the measurement precision of power consumption decreases
Solution Approach 1:
The patent divides the time period into multiple segments (first time period, second time period, etc.) and generates separate estimation models for each segment. This segmentation allows each model to specialize in specific time-range patterns, improving overall estimation accuracy without requiring a single overly complex model to handle all variations.
Solution Approach 2:
The patent implements dynamic model selection where the appropriate estimation model is chosen based on the current elapsed time. As time progresses, different models are applied at different stages, making the system adaptable to changing power consumption patterns rather than using a static single model.
2Measurement precision
If multiple estimation models are generated for different time periods, then the measurement precision of power consumption improves, but the device complexity increases
Solution Approach 1:
The patent pre-generates multiple estimation models in advance for different time periods before actual power consumption estimation is needed. This preliminary preparation allows the system to have ready-to-use specialized models, reducing the computational burden and complexity during real-time operation while maintaining high accuracy.
3Measurement precision
If power consumption is estimated for each job separately, then the measurement precision improves, but the loss of time increases
Solution Approach 1:
The patent combines multiple estimation results from different time period models to obtain the total power consumption for a job. By merging the estimates from the first time period model and second time period model, the system achieves comprehensive job-level accuracy while leveraging the efficiency of segmented processing.
Data Source
AI summary
A power estimation method includes acquiring first consumed power information indicating a change in power consumed for a first time period from a first time to a second time within a time period elapsed from the start of the execution of a first job, and second consumed power information indicating a change in power consumed for a second time period from a third time to a fourth time within the time period elapsed from the start of the execution of the first job, the third time being after the second time, generating, from the first consumed power information, a first estimation model for estimating power to be consumed by the job for the first time period, and generating, from the second consumed power information, a second estimation model for estimating power to be consumed by the job for the second time period.