Shrunk Dynamic Call Graphs for Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting anomalies in software applications are manual and time-consuming, leading to high Mean Time To Repair (MTTR) due to the expense of generating detailed dynamic call graphs in production environments.
Innovation Solution
A processor-implemented method using shrunk dynamic call graphs and machine learning to automatically identify anomalies by generating shrunk dynamic call graphs with a shrink factor, identifying anomalous groups, and determining root causes based on deviations and exceptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If detailed dynamic call graphs are generated to enable accurate anomaly detection, then measurement precision is improved, but loss of time and productivity deteriorate due to the expensive generation process
Solution Approach 1:
The patent extracts only the essential and meaningful information from detailed dynamic call graphs by identifying and retaining significant method calls and their relationships, while filtering out redundant or less important call details. This extraction process maintains the ability to detect anomalies accurately while reducing the time and computational resources required for graph generation and analysis.
2Measurement precision
If detailed dynamic call graphs are generated to capture meaningful execution information, then measurement precision is improved, but device complexity and resource consumption worsen due to huge memory and CPU requirements
Solution Approach 1:
The patent extracts and retains only the meaningful method calls and their hierarchical relationships from the dynamic call graph, filtering out redundant information. This selective extraction maintains execution analysis accuracy while significantly reducing memory and CPU requirements for processing and storing call graph data in production environments.
Solution Approach 2:
The patent applies different levels of detail and analysis to different parts of the call graph based on their significance. Critical method calls and anomaly-prone regions are analyzed in greater detail, while less important regions are summarized or aggregated. This local quality approach optimizes resource utilization by concentrating computational efforts where they are most needed.
3Measurement precision
If manual interpretation methods are used for anomaly detection, then measurement precision can be maintained through expert analysis, but productivity deteriorates due to the time-consuming manual process
Solution Approach 1:
The patent implements automated anomaly detection systems that self-analyze the extracted call graph information using machine learning models and statistical methods. The system automatically identifies anomalies, determines their severity, and suggests root causes without requiring manual interpretation, thereby maintaining detection accuracy while dramatically improving productivity and reducing Mean Time To Repair.
Data Source
AI summary
A system and method for real-time or near real-time anomaly detection and root cause automation in production environments or in other environments using shrunk dynamic call graphs are provided. The system includes an instrumentation agent that generates shrunk dynamic call graphs and exceptions/errors by injecting monitoring code or probes or call-tags into monitored application, a data agent that forwards collected data to the analysis engine over a network, an analysis engine that performs continuous clustering using machine learning, anomaly, and root cause detection. The system also includes a reporting module to report the anomaly.