Time-Series Correlation for Faster Root Cause Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods are inadequate for quickly and efficiently identifying candidate time series with a high probability of causing an event in a reference time series, particularly in cloud computing environments where numerous connections and data sources need to be analyzed.
Innovation Solution
A computer-implemented method involving correlation coefficient calculations, differentiation, shifting, and smoothing of time series data to identify candidate time series with a high correlation to a reference time series, using techniques like Pearson correlation, Savitzky-Golay filtering, and simplified time series construction based on event detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple candidate time series are analyzed to find root cause, then the accuracy of root cause identification is improved, but the computational complexity and time required increases
Solution Approach 1:
The patent segments the analysis process into distinct stages: preprocessing (smoothing, differentiation), correlation calculation, and candidate selection. Each stage processes the time series data independently, allowing the system to handle multiple candidate time series without exponentially increasing complexity. The segmentation enables parallel processing of different candidate series against the reference series.
Solution Approach 2:
The patent transforms the time series data through parameter changes including smoothing (changing the temporal resolution), differentiation (changing the derivative order), and correlation coefficient calculation (changing the statistical parameter). These transformations convert raw data into features that are more suitable for comparison, improving accuracy while maintaining manageable computational requirements through standardized operations.
2Measurement precision
If multiple candidate time series are analyzed to find root cause, then the accuracy of root cause identification is improved, but the time required for analysis increases
Solution Approach 1:
The patent performs preliminary actions by pre-processing the time series data before correlation analysis. Smoothing and differentiation are applied in advance to all candidate time series, and these transformed series are stored for subsequent comparison. This preliminary preparation eliminates the need to reprocess raw data for each candidate, significantly reducing the time required when analyzing multiple candidates.
Solution Approach 2:
The patent creates copies of the transformed time series data (smoothed and differentiated versions) that can be reused across multiple correlation calculations. Instead of processing原始数据 repeatedly, the system uses pre-computed copies, reducing redundant computations and accelerating the analysis of multiple candidate time series against the reference series.
3Measurement precision
If noise is reduced through smoothing, then the correlation accuracy is improved, but the computational operations increase
Solution Approach 1:
The patent extracts the essential signal characteristics from noisy time series data through smoothing operations. By separating the signal component from the noise component, the system retains only the relevant information needed for correlation analysis. This extraction process reduces the impact of noise on correlation accuracy while the standardized smoothing algorithms keep computational overhead manageable.
Data Source
AI summary
The disclosure concerns the time-based correlation of time series for the root cause analysis of application monitoring data, observability data, and data observability data in the field of application monitoring and observability. The object of the disclosure is to find at least one candidate time series that has a high chance/probability for causing an event in a reference time series. The method shall be time-based and not frequency based. The method includes: constructing a simplified reference time series retaining changes in the reference time series and setting other points to zero; constructing a simplified candidate time series retaining changes in the candidate time series and setting other points to zero; calculating a similarity metric between the simplified candidate time series and the simplified reference time series; and reporting an occurrence of a computing event in response to the similarity metric exceeding a threshold value.