Application Flow And Index Vectorization For Unified Root Cause Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for flow analysis in distributed systems require high repeated construction costs due to separate models for tasks like anomaly detection and root cause analysis, and lack efficient vector representation for network flows.
Innovation Solution
A method and system for application flow and index vectorization that quantizes network flows and index information into tuples, forms networks using time windows and random walks, and applies embedding algorithms to generate vectors for root cause analysis and prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate flow analysis models are built for different targets (anomaly detection, root cause analysis, flow prediction), then task-specific analysis accuracy is improved, but computing resource consumption and repeated construction cost increase
Solution Approach 1:
The patent builds a unified flow analysis model that can perform multiple tasks (anomaly detection, root cause analysis, flow prediction) simultaneously. The system constructs a single flow graph from quantized flow information that serves as a universal foundation for all analysis types, eliminating the need to separately build and maintain multiple specialized models. This universal model achieves task-specific accuracy through different query and analysis approaches on the same underlying structure.
Solution Approach 2:
The patent merges multiple flow analysis functions into a single integrated system. By combining anomaly detection, root cause analysis, and flow prediction capabilities into one unified model that shares common data structures and processing mechanisms, the system reduces repeated construction costs and computing resource consumption while maintaining the analytical depth needed for each specific task.
2Loss of information
If network flows are represented in detailed original form, then flow information completeness is improved, but processing efficiency and storage requirements worsen
Solution Approach 1:
The patent transforms network flow data by changing its parameter representation. Original flow information is quantized into discrete tuples with specific parameters (flow ID, source, destination, protocol, port, etc.). This parameter transformation converts continuous, high-volume network data into discrete, structured tuples that are more efficient to process and store while retaining essential flow characteristics needed for analysis.
Solution Approach 2:
The patent segments network flow information into discrete, structured tuples. Each flow is broken down into fundamental components (source, destination, protocol, ports, etc.) represented as separate fields in a tuple structure. This segmentation organizes complex flow data into manageable, standardized units that improve processing efficiency and reduce storage requirements while preserving the complete flow information needed for comprehensive analysis.
3Adaptability or versatility
If multiple flow analysis models are constructed for different tasks, then comprehensive coverage of analysis needs is improved, but system complexity and maintenance difficulty increase
Solution Approach 1:
The patent creates a universal flow analysis model that provides comprehensive coverage of different analysis needs through a single system. The unified model supports anomaly detection, root cause analysis, flow prediction, and other analysis tasks by using different queries and analysis approaches on the same underlying flow graph structure, thereby maintaining versatility while reducing system complexity.
Solution Approach 2:
The patent uses template-based flow graph construction where the same structural pattern is copied and applied across different analysis scenarios. Instead of building custom models for each task, the system copies and adapts the same fundamental flow graph structure and processing mechanisms, reducing maintenance difficulty and system complexity while maintaining comprehensive analytical coverage.
Data Source
AI summary
An inference-oriented method for application flow and index vectorization includes: S1: quantizing original index information of related devices with known network flows and current flows in a distributed system into tuples of a same type to obtain known network flow information tuples and index information tuples; S2: quantizing unknown network flows into tuples of a same type to obtain unknown network flow information tuples; S3: forming a network according to a time window and a priori association relationship based on the known network flow information tuples, the index information tuples, and the unknown network flow information tuples, and sampling network description information by means of random walk; S4: obtaining network flow vectors and index information vectors by means of an embedding algorithm based on the network description information; and S5: performing root cause analysis and index prediction according to the network flow vectors and the index information vectors.