Event Processing Operators for Stream Filtering and Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional database systems and business data processing algorithms are not well-equipped to handle stream-oriented applications with continuous data streams and long-running queries, as they prioritize current data states and treat notifications and triggers with low priority, assuming synchronized data and exact query answers, which is inadequate for applications like network monitoring and sensor networks.
Innovation Solution
The development of novel event processing operators for filtering, partitioning, abstraction, and estimation, which can be implemented in software or hardware, to process event streams in real-time by clustering, classification, hypothesis testing, and point estimation, allowing for the detection and handling of events outside predefined clusters or classifications, and the estimation of parameters for predicate evaluation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional database systems are used to store and process data, then current data state can be easily obtained, but previous values can only be found by decoding database log files and the system is not suitable for continuous data streams
Solution Approach 1:
The patent transforms the static database model into a dynamic event processing model where data flows continuously as events through a network of operators. Instead of querying a static repository, the system processes streams of events in real-time, allowing both historical and current data to be accessed through the event stream architecture.
Solution Approach 2:
The patent replaces the traditional database query mechanism with an event processing network using CQL operators. This substitution enables the system to handle continuous data streams and compute answers with incomplete information, rather than requiring complete data sets before processing.
2Ease of operation
If conventional database systems prioritize current data state, then current values are easy to obtain, but the system treats notifications and triggers with low priority and is not designed for asynchronous data
Solution Approach 1:
The patent inverts the conventional approach by making event processing and notifications the primary mechanism rather than an afterthought. The event processing network is designed from the ground up to handle asynchronous events with high priority, using operators that can process events in real-time and generate notifications immediately when conditions are met.
3Measurement precision
If conventional systems assume synchronized data and exact query answers, then queries have precise results, but the system cannot handle incomplete information in stream-oriented applications
Solution Approach 1:
The patent implements operators that can compute query answers with partial or incomplete information. Rather than waiting for complete data sets, the system processes events as they arrive and generates intermediate results, allowing queries to return best-effort answers based on available data while continuing to process incoming events.
4Productivity
If event processors use periodic cluster redefinition based on historical data, then events outside clusters can be filtered, but the system must balance processing speed with computational overhead
Solution Approach 1:
The patent implements periodic redefinition of event clusters based on historical data analysis. The system operates in cycles, where clusters are defined based on past event patterns and then used for filtering incoming events. This periodic approach balances the need for adaptive cluster definitions with the computational overhead of recalculating clusters, maintaining processing efficiency while adapting to changing event patterns.
Data Source
AI summary
Novel event-processing operators are provided. These novel operators can be advantageously utilized in implementing event processors and event-processing systems. The novel event-processing operators include operators for event filtering by clustering, operators for event partitioning by classification, operators for event abstraction by hypothesis testing, and operators for event filtering by point estimation.