Workload Power Estimation via Segmented Regression Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating workload power consumption, such as single linear regression models, tend to overestimate power consumption for low-power workloads when trained on datasets with a long tail of high-power data points, leading to inaccurate scheduling and resource utilization decisions.
Innovation Solution
A power consumption estimation system that trains multiple short-range power models and a classifier to estimate power consumption by dividing the entire power range into predetermined sections, using metrics like CPU, memory, and network usage to accurately predict power consumption within specific ranges, thereby preventing overestimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single linear regression model is trained on a data set with the full range of power consumption including high-power data points, then the model can cover the entire power range, but it overestimates the power consumption of low-power workloads due to the intercept being influenced by high-power data
Solution Approach 1:
The patent divides the entire power consumption range into multiple short ranges (e.g., first short range, second short range, third short range) and trains separate linear regression models for each range. This segmentation allows each model to be trained on data specific to its power range, preventing the intercept from being skewed by high-power data points when estimating low-power workload consumption. The classifier then selects the appropriate model based on the workload's power range.
2Measurement precision
If multiple short-range power models are trained to improve estimation accuracy for specific power ranges, then low-power workload estimation accuracy improves, but the system complexity increases due to multiple models and a classifier
Solution Approach 1:
The patent segments the power consumption range into multiple short ranges with separate models, which improves accuracy but increases complexity. This segmentation is necessary to resolve the overestimation problem while maintaining manageable model sizes.
Solution Approach 2:
The patent introduces a classifier as an intermediary component that receives server metrics as input and outputs specification information identifying which short-range power model should be used. This classifier acts as a mediator between the input metrics and the multiple power models, automatically selecting the appropriate model based on the current power range, thereby managing the complexity of having multiple models.
Data Source
AI summary
To estimate power consumption of a workload, an estimation server estimates power consumption of a workload executed on a physical server. A processor trains a plurality of short-range power models that receive a metric of the physical server as an input and output a power consumption value of the physical server in a plurality of short ranges obtained by dividing an entire power range of the physical server into a predetermined division number, trains a classifier that receives a metric of the physical server as an input and outputs specification information specifying a corresponding short range and specifies specification information specifying a short range to be applied based on a metric of the workload and the classifier. The estimate of the power consumption of the workload is based on the metric of the workload and a short-range power model corresponding to the short range indicated by the specification information.