Cloud Workload Energy Estimation via Multi-Horizon Regression Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for calculating energy consumption in cloud computing environments are inadequate for volatile workloads with fluctuating processing demands, as they are designed for fixed patterns and lose accuracy when workload patterns change.
Innovation Solution
A method that collects periodic samples of system activities and node energy consumption levels to build short-term and long-term regression models, allowing for accurate estimation of energy consumption by capturing both short-term and long-term workload patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a single fixed pattern model is used for energy consumption calculation, then the method is simple to implement, but the accuracy deteriorates when workload patterns change
Solution Approach 1:
The patent segments the energy consumption estimation into multiple models with different time horizons (short-term and long-term). Each model handles specific temporal patterns separately, allowing the system to maintain accuracy across varying workload conditions without requiring a single complex universal model.
Solution Approach 2:
The patent implements dynamic model selection where the system adapts between short-term and long-term models based on workload characteristics. This dynamic approach allows the estimation accuracy to adjust automatically to changing conditions while maintaining implementation simplicity through automated model switching.
2Measurement precision
If multiple models with different time horizons are used, then the energy consumption estimation accuracy improves for volatile workloads, but the device complexity increases
Solution Approach 1:
The patent creates a universal estimation system where multiple models serve different time horizon requirements within a single integrated framework. The system manages both short-term and long-term models through unified mechanisms, allowing one system to handle multiple estimation scenarios without proportionally increasing complexity.
Solution Approach 2:
The patent incorporates feedback mechanisms that monitor workload patterns and automatically select between models based on observed conditions. This feedback-driven approach simplifies model management by automating selection decisions, reducing the operational complexity that would otherwise arise from manually managing multiple models.
Data Source
AI summary
Computer-implemented methods for estimating energy consumption of a workload in a cloud computing system are provided. Aspects include periodically collecting an energy consumption data for the workload, creating a first model based on the energy consumption data corresponding to a first duration, and creating a second model based on the energy consumption data corresponding to a second duration, wherein the second duration is longer than the first duration. Aspects also include receiving a request for an estimated energy consumption of a workload during a time period and calculating a first estimated energy consumption of the workload during the time period based on the first model. Aspects further include calculating a second estimated energy consumption of the workload during the time period based on the second model and calculating a combined estimated energy consumption of the workload based on the first estimate and the second estimate.