- The paper presents Text2Net, a system that translates plain-text network descriptions into dynamic simulations using NLP and LLMs, automating configuration.
- Quantitative evaluation shows Text2Net drastically cuts down configuration time and steps for network setups compared to manual methods.
- Text2Net enhances efficiency and usability for network prototyping and education, offering a valuable tool with planned expansions for complex network configurations.
The paper presents a system that translates plain-text descriptions of network topologies into a dynamic, interactive network simulation environment. The approach leverages advances in NLP and large LLMs to automate the generation of network configurations, reducing the reliance on vendor-specific syntaxes and complex graphical interfaces.
The proposed system architecture consists of five key modules:
- User Module: Accepts plain-text input describing network scenarios.
- Software-Adaptor: Acts as an intermediary, forwarding input to the instructed LLM and subsequently processing the returned Structured Command Strings (SCS).
- Instructed LLM: A tailored version of ChatGPT-4T that translates textual descriptions into precise command strings.
- Text Extractor: Utilizes NLP tools (e.g., SpaCy), RegEx, and pattern matching to extract key-value pairs and constructs a JSON representation of the configuration.
- Simulator: Integrates with the EVE-NG emulator to provision the network topology based on the structured JSON configuration.
The paper describes the complete workflow from user input to network deployment. After receiving a network description in plain English, the system validates the input and extracts essential parameters such as device types, interfaces, IP addresses, and routing details. An algorithm is detailed that iterates over segmented SCSs, extracts attributes (e.g., node type, node name, interface details), assigns unique identifiers, and consolidates connection and routing information into a JSON blueprint. This blueprint is then used to dynamically provision devices and interconnections in the EVE-NG emulation environment.
Key technical aspects include:
- Input Standardization and Validation: The system incorporates robustness features to handle variations in natural language descriptions. It prompts for additional details when critical information, such as static routing parameters, is missing—for example, invalid IP address detection and incomplete configuration details generate user prompts for clarification.
- Algorithm for Data Extraction: A pseudocode algorithm systematically dissects the plain text into key-value pairs, effectively mapping user input to device-specific configurations. It uses functions such as ExtractNode, ExtractName, and AssignUniqueID to ensure each network component is uniquely identified and correctly configured.
- Integration with Emulation Environment: The JSON output serves as a blueprint for creating virtual nodes (e.g., routers, switches) and linking them according to the provided topology. The dynamic creation of nodes, network links, and subsequent configuration via API interactions demonstrates a significant reduction in manual configuration steps.
- Quantitative Evaluation: Three network scenarios of increasing complexity were used to benchmark performance. For instance:
- In Scenario 1, the manual configuration took approximately 200 seconds over 12 steps, while Text2Net reduced this to 110 seconds in just 2 steps.
- In Scenario 2, involving two interconnected routers and additional loopback interfaces, the manual configuration required 510 seconds compared to 250 seconds using the system.
- In Scenario 3, which introduced a third router and additional static routing complications, the manual method required 730 seconds versus 310 seconds with the proposed approach.
- Qualitative Assessment: Feedback from graduate students, faculty, and industry practitioners yielded an average rating of 4.66 out of 5 regarding ease of use, error reduction, and educational value. The system’s ability to significantly reduce repetitive tasks and streamline complex configurations is emphasized.
Future directions outlined in the paper include extending the system to support Layer 2 protocols (e.g., VLAN and Spanning Tree Protocol) and advanced configurations (e.g., NAT and VPN). The authors propose replacing the current RegEx and pattern-matching mechanisms with robust NLP techniques such as LangChain and RAGs (Retrieval Augmented Generators), aiming to enhance scalability and accuracy in generating network commands.
In summary, the paper details a comprehensive framework for automating network simulation through plain-text input, significantly reducing the operational overhead associated with traditional manual configuration tasks. The system presents substantial improvements in efficiency and usability, positioning it as a valuable tool for both network education and professional prototyping environments.
