Google Gemini for Network Configuration Assistance

This blog post explores how Google Gemini, a generative AI model, can enhance efficiency and productivity for network engineering and provisioning teams. It focuses on a scenario involving two retail sites connected through a Telco-managed L3 VPN service. The blog outlines how Gemini can assist in IP address planning, network topology summarization, configuration script generation, and documentation, ultimately streamlining the delivery of L3 VPN services. The author emphasizes Gemini’s potential to reduce manual processes and accelerate network automation, while acknowledging the need for human verification to ensure accuracy.

Introduction

IP network is integral and critical part of Telco’s network infrastructure ecosystem for L2 and L3 VPN (virtual private network) connectivity services for captive use (e.g. Mobile backhaul, own offices & retail sites connectivity) and/or offering these VPN services to their Enterprise customers for office connectivity.

Typically after the order has been closed by Telco enterprise sales teams, the enterprise customer requirements and scenario is passed to the network planning and provisioning teams for fulfillment of the order. Apart from creating customer specific configuration for the given scenario; documentation is a critical part of the service and these activities may take a few hours to days depending upon the number of the sites and PoP sites involved for delivering the service.

In this blog, we will be exploring how Generative AI (powered by multi-modal Google Gemini 1.5 flash) can be leveraged for assisting network engineering/provisioning teams to increase efficiency & productivity in terms of:

• IP addressing planning & calculations
• Summarization of the Network Topology and other aspects of the network
• Faster generation of configuration scripts/templates and documentation of the service
• Generating XML/JSON files for Automation

Multimodality (understanding of text, images, audio and video) and 1M context window aspects of the Gemini 1.5 Flash become quite useful in order to abstract the information from the Network topology and document it for the network engineers, apart from performing other activities.

Example Scenario

In the above scenario, Telco will not only be responsible for configuration of Core and Edge routers in its MPLS network, but will also be responsible for configuration & management of CPE (Customer Premise Equipment) routers since this is managed service offering. The two the sites connectivity may not seem to be a big concern, but in real large enterprise connectivity involving 10s & 100s of sites/Provider Edge routers/links etc. it becomes quite complex & time consuming from efficiency & productivity point of view.

User Journey

Delivery of L3 VPN service is a multi-step process as shown in the above example and is executed in manual or semi-automatic way depending on the level of automation done by a Telco. The applicability of the Gemini LLM as shown in the above figure is being considered in

• Step 3 [generating the /30 IP addresses based on the pool of class C provided by IPAM system for various links/interfaces and documenting in a tabular fashion all the IP addresses being used]
• Step 4 [mapping of all different IP addresses as calculated in step 3 to the network topology and create configuration & verification scripts for all the PE and CE routers]
• Step 5 [Documentation of all Service for future reference including explanation of Network Scenario, IP addressing scheme, Configuration files & steps, Verification steps etc.
• Further there can be steps of generating JSON/NETCONF xml files for each router configuration to enable automation

All of the above steps generally involve manual/semi-automated processes, working through multiple sheets, config templates etc (unless Telco has achieved 100% automation already) and may take a few hours to a few days depending on the scale of network & connectivity.

Next we will see how Gemini powered assistants can help the Network engineer simplifying these steps reducing time and increasing productivity.

Leveraging Gemini

1. Calculation of IP Addresses & Summarization Table (<10 Sec)

Systems Instructions for the Assistant <example>

Input Prompt for calculating required /30 IP addresses along with summary table & explanation

2. Network Image analysis & Information Abstraction (<10 Sec)

After the IP addresses calculation, network engineers will be mapping the IP addresses to various links and interfaces as per the design requirements. Once the mapping is done, the network engineer will upload the image as input prompt and based on system instructions Gemini assistant will provide the summary explanation and table of all the IP addresses, AS number, Site names etc.

3. Generating Network Configurations & Verification Scripts (<45 Sec)

As next step, based on the input prompt instructions as shown, the assistant will generate the configuration & verification scripts for each of the Core & Edge routers, CE routers along with all the steps followed for easy documentation. Although LLM provides a good start with substantial accuracy for the network engineer to have quickly generated config templates, he/she recommended to still verify the configuration scripts for any missteps or correctness against golden config templates. This step can be further automated however not captured while writing this blog.

4. Generating NETCONF XML files for Automation and Summarization of Steps (~45 Sec)

Summary

As we saw in above examples, how multi-modal Gemini can be used for abstraction and summarization from the network image as well as generating configuration scripts/verification scripts. We also experienced how generative AI can be leveraged for assisting network engineering/provisioning teams to increase efficiency & productivity reducing hours/days of work into minutes, conversion of configuration files into XML formats, and boosting the network automation. The accuracy can be further improved by fine tuning the model by implying multi-shot prompting.

Disclaimer: This is to confirm readers that the views, thoughts, and opinions expressed in the text belong solely to the author, and not necessarily to the author’s employer, organization, committee or other group or individual.