Probe Deployment for Cloud Application Performance Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud application deployment on virtual machines hosting existing cloud applications incurs high overhead and complexity, making it difficult to predict performance without deploying the actual new application, which is resource-intensive and complex.
Innovation Solution
A probe is generated and deployed into a virtual machine to increase resource consumption incrementally, collect performance data, and train a machine learning model to predict cloud application performance without the need for deploying the new application, thereby reducing resource usage and complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a new cloud application is deployed to predict performance, then performance data can be collected, but resource consumption and overhead increase significantly
Solution Approach 1:
The patent creates a simplified probe that copies only the essential resource consumption characteristics of a new cloud application without including the full application functionality. This probe serves as a lightweight replica that mimics resource usage patterns (CPU, memory, I/O) while being significantly smaller and less resource-intensive than the actual application, enabling performance prediction without deploying the complete application stack
Solution Approach 2:
The probe is designed as a temporary, disposable artifact that is deployed only for the purpose of collecting performance data and then removed. It is not a permanent part of the system but rather a transient tool used to gather information about resource consumption patterns under different load conditions, after which it is discarded and the actual application deployment proceeds
2Measurement precision
If a new cloud application is deployed to test performance, then accurate performance data is obtained, but deployment complexity increases
Solution Approach 1:
Instead of deploying the complex full application stack for testing purposes, the patent deploys a simplified probe that copies only the necessary resource consumption behavior. This probe is much simpler to deploy, configure, and manage while still providing the performance data needed for prediction, thereby reducing deployment complexity significantly
Solution Approach 2:
The patent separates the performance testing function from the actual application deployment by creating a distinct probe component. This segmentation allows the probe to be deployed independently for performance characterization, with its results used to predict application behavior, thus avoiding the complexity of deploying and managing the full application just for testing purposes
3Measurement precision
If resource consumption by the probe is increased to improve prediction accuracy, then performance data at higher loads is collected, but overhead on the virtual machine increases
Solution Approach 1:
The patent implements dynamic resource consumption adjustment in the probe, allowing it to operate at multiple resource consumption levels. The probe can be configured to consume resources at different intensities (low, medium, high) depending on the prediction scenario, enabling it to simulate various load conditions on the virtual machine and collect performance data across different operating points without permanently committing to high resource usage
Data Source
AI summary
A method includes generating, by a computing device, a probe; deploying, by the computing device, the probe into a virtual machine hosting a cloud application; increasing, by the computing device, resource consumption by the probe; collecting, by the computing device, performance data from the cloud application; and training, by the computing device, a machine learning model using the performance data.