Incident Dependency Prediction for Faster Network Root Cause Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cloud computing environments face inefficiencies in identifying responsible incidents and their dependencies during service disruptions, leading to prolonged downtime and reduced reliability due to manual investigation methods.
Innovation Solution
An automated incident investigation system using vector-based feature analysis and an incident dependency prediction engine trained on historical data to swiftly identify and predict root cause incidents, leveraging graph neural networks or instruction-tuned graph language models for intelligent dependency prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual investigation methods are used to identify responsible incidents, then operational simplicity is maintained, but service downtime increases and reliability decreases
Solution Approach 1:
The patent replaces manual investigation methods (mechanical human operation) with an automated incident dependency prediction system that uses machine learning models, vector-based feature analysis, and graph neural networks to automatically identify responsible incidents and their dependencies, thereby reducing service downtime and improving reliability
Solution Approach 2:
The system performs preliminary analysis by pre-processing incident data, building dependency graphs, and training prediction models in advance so that when an incident occurs, the system can rapidly query and predict the responsible incident using pre-computed relationships and historical patterns
2Productivity
If manual investigation methods are used, then system complexity is low, but productivity in incident identification decreases
Solution Approach 1:
The patent substitutes manual investigation processes with an automated computational system that leverages machine learning models, vector embeddings, and graph neural networks to rapidly identify incidents, achieving high productivity through algorithmic automation despite increased system complexity
Solution Approach 2:
The system creates vector-based copies of incident features and historical incident patterns, allowing rapid comparison and matching without processing raw data each time, thereby improving identification speed while managing complexity through efficient data representation
3Productivity
If automated prediction models are deployed, then incident identification speed increases, but measurement precision requirements increase
Solution Approach 1:
The patent transforms incident data into vector representations with multiple features and dimensions, allowing the prediction model to analyze incidents from multiple angles simultaneously, which improves both identification speed and precision by capturing nuanced relationships in the data
Solution Approach 2:
The system incorporates feedback mechanisms where prediction results are continuously refined using historical incident data and model performance metrics, allowing the system to learn from past predictions and improve measurement precision over time while maintaining high identification speed
Data Source
AI summary
A disclosed method for incident dependency prediction includes identifying, for a newly-occurring network incident of interest, one or more similar historical incidents; obtaining causal dependency data for the one or more similar historical incidents, the causal dependency data identifying one or more historical responsible incidents responsible for causing the similar historical incidents; identifying candidate root cause incidents each occurring within a recent time interval and satisfying similarity criteria with a corresponding one of the one or more historical responsible incidents; and transmitting a prompt to an incident dependency prediction model, the prompt requesting prediction of a root cause incident responsible for causing the newly-occurring network incident of interest based on feature information for each of the candidate root cause incidents and for the incident of interest.