Metric Ranking by Differential Analysis for Fault Prioritization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in network monitoring is to efficiently analyze vast amounts of telemetry data to identify significant metric changes and prioritize them for root cause analysis, as visual inspection is error-prone and time-consuming due to the large volume of metrics.
Innovation Solution
A system that automatically ranks metrics by their differences between reference and query periods using methods like cross-correlation, Dynamic Time-Warping, and Kolmogorov-Smirnov tests, combined through linear or nonlinear models to provide a quantifiable estimate of difference magnitude, facilitating the identification of anomalous performance and guiding deep dive analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual inspection of metrics is performed, then human operators can detect metric changes, but the process becomes error-prone and time-consuming due to large volume of metrics
Solution Approach 1:
The patent replaces manual visual inspection with automated computational methods including cross-correlation analysis, Dynamic Time-Warping, and Kolmogorov-Smirnov tests. These algorithms objectively compare metric time series data between reference and query periods, eliminating human error and significantly reducing analysis time while maintaining high detection accuracy for subtle metric changes
Solution Approach 2:
The system performs self-diagnosis by automatically detecting anomalies and ranking metrics without human intervention. The automated anomaly detection system independently analyzes telemetry data, identifies significant changes, and prioritizes metrics for investigation, enabling the monitoring system to serve itself rather than relying on manual operational analysis
2Measurement precision
If metrics are analyzed one by one, then subtle differences can be detected more easily, but it becomes difficult to rank metrics by strength of relative changes and very time consuming
Solution Approach 1:
The patent segments the analysis process into distinct computational stages: first applying multiple statistical tests (cross-correlation, DTW, KS-test) to individual metric pairs, then aggregating results through linear or nonlinear combination models, and finally ranking metrics by composite anomaly scores. This segmented approach maintains precise detection capability while enabling batch processing of numerous metrics simultaneously
Solution Approach 2:
The system merges multiple statistical test results into a unified anomaly detection framework. By combining cross-correlation coefficients, DTW distances, and KS-test statistics through weighted linear combinations or nonlinear models, the patent synthesizes individual metric analyses into a comprehensive ranked list, achieving both detection precision and high throughput
3Loss of time
If automated ranking of metrics by difference magnitude is implemented, then Time-To-Resolve is reduced, but complex computational models are required to quantify difference magnitude
Solution Approach 1:
The patent introduces intermediary statistical measures that bridge raw metric data and final anomaly rankings. Cross-correlation coefficients, DTW distances, and KS-test statistics serve as intermediate representations that quantify differences between metric periods. These intermediaries are then combined through relatively simple linear or nonlinear models to produce final rankings, reducing computational complexity compared to direct complex optimization approaches
Data Source
AI summary
The technology disclosed relates to differential analysis of sets of time series pairs. In particular, it relates to building estimators of magnitude of difference between two time series. After the basic estimators are built, they are combined into ensemble estimators using linear or nonlinear prediction models to improve their accuracy. In one application, the ensemble is used for estimating the magnitudes of difference over sets of metric pairs observed from distributed applications and systems running over a computer network. The metric pairs are then ranked in decreasing order of magnitude of difference to guide an operator in prioritizing his root cause analysis of faults, thereby reducing the time-to-resolution of problems.