Pattern Analytics for Real-Time Failure Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to accurately predict and prevent catastrophic failures in cloud computing systems, as they often occur suddenly and are difficult to pinpoint, leading to costly repairs and data loss, with existing methods unable to reliably identify known significant pattern signatures that may precede such failures.
Innovation Solution
A method using a processor to generate a candidate pattern list from historical data by sweeping sliding windows and applying fuzzy logic to determine correlations, identifying maximal cliques as significant patterns, and predicting causal relationships between patterns to alert administrators in real-time, allowing for potential failure mitigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing monitoring systems are used to detect system failures, then system operation continues without interruption, but catastrophic failures occur suddenly and cannot be predicted
Solution Approach 1:
The system performs preliminary actions by continuously analyzing historical data to identify and store significant pattern signatures before failures occur. The method sweeps through historical data with sliding windows to extract patterns, stores them in a library, and then uses these pre-established patterns to predict future failures, enabling proactive rather than reactive failure management.
Solution Approach 2:
The system performs preliminary actions by continuously analyzing historical data to identify and store significant pattern signatures before failures occur. The method sweeps through historical data with sliding windows to extract patterns, stores them in a library, and then uses these pre-established patterns to predict future failures, enabling proactive rather than reactive failure management.
2Measurement precision
If traditional pattern matching methods are used, then simple patterns can be detected, but known significant pattern signatures cannot be reliably identified
Solution Approach 1:
The system segments the complex task of failure prediction into distinct components: (1) sweeping historical data with sliding windows to generate candidate patterns, (2) using fuzzy logic to determine correlations and identify maximal cliques, (3) storing significant patterns in a library, and (4) matching real-time data against the library. This segmentation enables reliable identification of significant patterns while managing system complexity through modular processing.
Solution Approach 2:
The system introduces an intermediary pattern library that stores pre-identified significant patterns. This library acts as a mediator between historical data analysis and real-time failure detection, enabling accurate pattern matching without requiring complex real-time analysis of all historical data, thus improving detection accuracy while controlling complexity.
3Measurement precision
If comprehensive historical data analysis is performed, then accurate pattern signatures can be identified, but processing time and computational resources increase
Solution Approach 1:
The system performs comprehensive historical data analysis in advance to build a library of significant pattern signatures. By sweeping through historical data with sliding windows and using fuzzy logic to identify maximal cliques beforehand, the system prepares prediction patterns ahead of time, enabling fast real-time detection without repeating the full analysis process.
Solution Approach 2:
The system extracts only the most significant patterns from comprehensive historical data analysis by identifying maximal cliques and storing them as signature patterns in a library. This extraction process separates the essential predictive patterns from the vast amount of historical data, enabling accurate failure prediction with minimal processing requirements during real-time operation.
Data Source
AI summary
A method includes reading a plurality of historical data from a storage. The method also includes generating, using a processor, a candidate pattern list. The method also includes determining, using the points stored in the candidate pattern list, whether at least one maximal clique exists in the candidate pattern list, each maximal clique indicating a known significant pattern. The method further includes, in response to determining that at least one maximal clique exists in the candidate pattern list, for each of the at least one maximal cliques, determining a signature of the maximal clique and storing the signature of the maximal clique and an identifier identifying the known significant pattern to a library in the storage. The method further includes determining causal relationships between the identified significant patterns and storing in the library.