ML Network Traffic Generation for Data Center Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current network testing methods fail to effectively mimic real-world scenarios in data center environments, lacking the ability to emulate switching fabrics and other resources accurately, which is crucial for thorough testing of network systems.
Innovation Solution
The implementation of a machine learning-based network traffic generation system that collects live traffic from production data centers and emulated testbeds, trains a traffic generation inference engine, and generates test traffic to stimulate network systems under test, utilizing emulated switching fabrics and virtualized resources to mimic real-world conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional network testing methods are used, then testing can be performed with simple setups, but the testing cannot accurately mimic real-world scenarios and conditions
Solution Approach 1:
The patent creates virtual copies of production data center environments, switching fabrics, and network resources within a testbed. These virtualized copies replicate the behavior and characteristics of real systems without requiring physical duplication, enabling realistic testing while managing complexity through software-based emulation
Solution Approach 2:
The testing system is divided into separate functional components: production environment collection, testbed emulation, machine learning training, and traffic generation. This segmentation allows each component to be developed and maintained independently while working together to achieve accurate real-world scenario replication
2Adaptability or versatility
If virtualization is used to emulate resources, then testing can be performed without physical equipment, but the emulation may not accurately approximate various equipment or system states
Solution Approach 1:
The system performs preliminary actions by collecting actual traffic data from production environments before creating test scenarios. This real traffic data is used to train machine learning models that understand genuine network behavior patterns, which then guide the virtualized testbed to accurately approximate real equipment and system states
Solution Approach 2:
The system incorporates feedback loops where test results from the virtualized testbed are analyzed and used to refine the emulation accuracy. Machine learning models are continuously trained on both production traffic and testbed traffic to improve the approximation of real-world equipment states and behaviors
Data Source
AI summary
Methods, systems, and computer readable media for network traffic generation using machine learning. An example method includes collecting first traffic from a production data center environment. At least a portion of the first traffic comprises live computer network traffic transiting the production data center environment. The method includes collecting second traffic from an emulated data center testbed device. At least a portion of the second traffic comprises testbed traffic that transits an emulated data center switching fabric of the emulated data center testbed device. The method includes training a traffic generation inference engine using the first traffic and the second traffic. The method includes generating, using the traffic generation inference engine, test traffic to test or stimulate a network system under test (SUT).