Time Series Analysis for Computing Resource Capacity Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for capacity management in computing systems are ineffective due to manual monitoring, failure to accurately detect change-points and trends in time-series data, leading to frequent outages and inefficient resource utilization.
Innovation Solution
A system that performs time-series analysis using machine learning-based prediction models to detect trend changes and level shifts in resource utilization data, enabling predictive maintenance and corrective actions such as reconfiguring networks or sending alerts to prevent outages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If manual capacity monitoring techniques are used, then system complexity is reduced, but capacity management reliability deteriorates due to frequent outages
Solution Approach 1:
The patent replaces manual capacity monitoring (mechanical/systematic approach) with automated machine learning-based time series analysis. The system uses prediction models to automatically detect change-points and trends in resource utilization data, eliminating the need for manual monitoring while significantly improving detection accuracy and preventing outages.
Solution Approach 2:
The system enables self-service capacity management by automatically analyzing time series data, detecting change-points, predicting future resource utilization, and generating alerts without human intervention. The machine learning models continuously learn from historical data and autonomously identify capacity issues before they cause outages.
2Use of energy by moving object
If conventional time series analysis techniques are used, then computational resources are saved, but measurement precision deteriorates due to inability to detect change-points accurately
Solution Approach 1:
The patent changes the parameters of time series analysis by using machine learning-based prediction models instead of conventional statistical methods. The system fits prediction models to historical data and analyzes residuals to detect change-points, achieving high detection precision while maintaining computational efficiency through optimized model selection and incremental updating.
3Ease of operation
If conventional statistical analysis techniques are used, then ease of operation is improved, but manufacturing precision deteriorates due to insufficient sample size requirements
Solution Approach 1:
The patent applies partial action by using only the necessary historical data samples required for training prediction models, rather than requiring hundreds of samples as conventional techniques demand. The machine learning models achieve accurate change-point detection with limited data by leveraging pattern recognition and iterative learning, maintaining ease of operation while improving prediction accuracy.
Data Source
AI summary
An online system receives time series data and analyzes the data for identifying trend changes or level shifts in the time series. The time series data may describe resource utilization of systems, for example, bandwidth of computer networks. The online system uses prediction models, for example, machine learning based prediction models using regression to predict data values for the time series. The online system determines error residue values based on difference between predicted data values and actual data values of the time series. The online system determines level change in the error residue to identify change-points representing trend changes or level-shifts in the original time-series. The online system takes corrective action based on the change-point information and the trend following the occurrence of the change-point, for example, sending alerts or instructions for causing reconfiguration of the systems such as the computer network.