Time Series Data Sampling for Cloud Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In cloud-based application acceleration as a service environments with numerous network entities, detecting and predicting data pattern changes is tedious and resource-intensive due to the large number of entities and associated features, making it challenging to identify problematic patterns in real-time.
Innovation Solution
A method and system that involves sampling time series data from each network entity into smaller intervals, generating reference data bands based on predicted future data sets, and detecting anomalies by comparing real-time data against these bands, using optimization algorithms to identify patterned anomalies across features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If time series data from all network entities is monitored in real-time with fine granularity, then detection precision of anomalies is improved, but computational cost and storage requirements increase significantly
Solution Approach 1:
The patent segments the continuous time series data into smaller time intervals and extracts only the maximum and minimum values from each interval. This segmentation approach reduces the data volume significantly while preserving the essential information needed for anomaly detection, thereby lowering computational cost and storage requirements while maintaining detection precision.
Solution Approach 2:
The patent extracts only the critical features (maximum and minimum values) from the complete time series data within each time interval. By taking out only the essential information needed for anomaly detection rather than processing the entire data set, the system achieves efficient anomaly detection with reduced computational overhead.
2Measurement precision
If data from all network entities is monitored continuously, then detection accuracy of problematic patterns is improved, but storage requirements increase significantly
Solution Approach 1:
The patent divides the continuous data stream into discrete time intervals and stores only the maximum and minimum values for each interval rather than the complete time series. This segmentation strategy maintains the ability to detect patterns and anomalies while dramatically reducing the quantity of data that needs to be stored.
Solution Approach 2:
The patent applies different data retention strategies to different aspects of the data: it stores maximum and minimum values for anomaly detection while not storing the complete intermediate values. This local quality approach ensures that the essential information for pattern detection is preserved while minimizing overall storage requirements.
3Measurement precision
If fine-grained time series data is processed for all features, then detection accuracy of anomalies is improved, but processing time increases
Solution Approach 1:
The patent processes data in segmented time intervals rather than continuously analyzing every data point. By dividing the time series into intervals and processing only the aggregated maximum and minimum values, the system maintains high detection accuracy while significantly reducing the time required for data processing.
Solution Approach 2:
The patent performs preliminary aggregation of time series data into maximum and minimum values before the actual anomaly detection process. This preliminary action reduces the data volume that needs to be processed in the main detection algorithm, thereby reducing overall processing time while maintaining detection accuracy.
Data Source
AI summary
A method implemented through a server of a cloud computing network including subscribers of application acceleration as a service provided therethrough includes sampling time series data associated with each network entity for each feature thereof into a smaller time interval as a first data series and a second data series including a maximum value and a minimum value respectively of the sampled time series data for the each feature within the smaller time interval, and generating a reference data band from predicted future data sets. The method also includes detecting, based on the reference data band, an anomaly in real-time data associated with the each network entity for the each feature thereof and determining an event associated with a pattern of change of the real-time data associated with the each network entity based on executing an optimization algorithm to determine a series of anomalies including the detected anomaly.