Dynamic Cloud Alert Generation Using Graph Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Predicting failures in complex cloud infrastructure is challenging due to the dynamic nature of software applications and the difficulty in understanding complex graphical patterns of events, with existing solutions relying on static threshold criteria that are ineffective in large, rapidly changing systems.
Innovation Solution
A dynamic alert and threshold generation system using causation mining with a graph neural network (GNN) that generates alerts and revises thresholds based on node classification, predicting failure impacts and adjusting thresholds dynamically to prevent faults.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If static threshold criteria are used for failure prediction, then the system is simple to implement, but it is ineffective in large, rapidly changing systems
Solution Approach 1:
The patent implements dynamic alert and threshold generation that adapts to changing system conditions. The system continuously learns from historical data and adjusts thresholds dynamically rather than using fixed static criteria, enabling effective failure prediction in rapidly changing cloud infrastructure environments
Solution Approach 2:
The patent introduces a graph neural network as an intermediary between raw monitoring data and failure predictions. The GNN processes complex graphical patterns of events and relationships, transforming them into actionable insights without requiring direct complex rule-based logic from the system operators
2Measurement precision
If complex graphical patterns of events are analyzed, then failure prediction accuracy improves, but the difficulty of detecting and measuring increases
Solution Approach 1:
The patent replaces manual or rule-based pattern analysis with a graph neural network that automatically processes complex graphical patterns. The GNN learns to detect relationships and patterns in event data without requiring explicit programming of analysis rules, significantly reducing the difficulty of detecting and measuring complex patterns
3Adaptability or versatility
If dynamic threshold adjustment is implemented, then adaptability to changing systems improves, but device complexity increases
Solution Approach 1:
The patent implements self-service through automated alert and threshold generation. The system automatically adjusts thresholds based on learned patterns from historical data and current system state, eliminating the need for manual threshold configuration and updating while maintaining high adaptability to changing conditions
4Reliability
If graph neural network classification is used for node classification, then failure prediction capability improves, but computational power requirements increase
Solution Approach 1:
The patent applies preliminary action by pre-training the graph neural network on historical data before deployment. The GNN learns failure patterns and relationships in advance, enabling it to make accurate predictions with reduced computational requirements during real-time operation compared to training models on-the-fly
Data Source
AI summary
Embodiments predict failures in a cloud infrastructure. Embodiments generate a graphical representation of a plurality of features of the cloud based network, the graphical representation including a plurality of nodes and corresponding relationships between the nodes, each node corresponding to one of the plurality of features. Embodiments monitor for events for the plurality of features, the events corresponding to one or more of the nodes, to generate monitored events, and populate a graph database with the monitored events. Embodiments classify each of the nodes with a trained graph neural network (“GNN”), the classification including a prediction of a failure of at least one node. Based on the classifying, for a first failure node corresponding to the prediction, embodiments generate a new alert corresponding to the first failure node.