Point-in-Time Outlier Detection for Sparse Metric Anomalies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing metric data in computing systems are inefficient and error-prone, especially when identifying sparse patterns of problematic behavior, which can lead to severe issues like slow system response times and outages.
Innovation Solution
A point-in-time outlier detection system using an ensemble of machine-learning models that analyze current and relative/absolute changes in metric data without pre-training, allowing for real-time detection of anomalous behavior in computing resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual analysis of log files and raw metric data is performed, then detailed investigation of system behavior is possible, but the process becomes inefficient and time-consuming
Solution Approach 1:
The patent replaces manual mechanical analysis of log files and metric data with automated machine learning models. The system uses unsupervised learning algorithms to automatically detect anomalies in computing resource behavior, substituting human analysts with computational systems that can process vast amounts of data rapidly while maintaining high detection accuracy for sparse problematic patterns.
2Reliability
If comprehensive metric data collection is implemented, then complete system behavior analysis is achieved, but data interpretation becomes increasingly difficult
Solution Approach 1:
The patent extracts only the most relevant features and patterns from comprehensive metric data using machine learning models. Instead of requiring analysts to interpret all collected metrics, the system automatically identifies and extracts meaningful anomalies and problematic patterns, presenting only the critical information that requires human attention while filtering out noise and redundancy.
3Productivity
If traditional outlier detection methods are used, then simple anomalies can be identified, but sparse patterns of problematic behavior are missed
Solution Approach 1:
The patent changes the detection parameters and thresholds dynamically based on learned patterns from historical data. The machine learning models adapt their sensitivity and detection criteria to identify sparse problematic patterns that traditional fixed-threshold methods would miss, while maintaining high-speed processing capability through efficient algorithmic implementations.
Data Source
AI summary
Techniques for performing point-in-time relative outlier detection are disclosed herein. In some embodiments, an outlier detection system analyzes metric data based on (a) the values of the metric data detected on a computing resource, (b) a relative change between different metric readings and/or (c) an absolute change between different metric readings relative to the point in time. The outlier detection system may predict whether the computing resource is exhibiting anomalous behavior by applying a set of machine-learning (ML) models to the point-in-time values. The ML models allow the outlook detection system to make inferences and adjustments during application runtime rather than relying on static instruction sets to detect and classify outliers. The ML models that are applied may implement unsupervised learning methods that do not rely on pre-training and/or time-series analysis for classification. Thus, the ML models may provide a point-in-time classification without requiring historical metric data to detect outliers.