Solutions Tech Lab

As you might have heard that Aruba Central was upgraded recently and now it is enhanced with Generative AI (GenAI) based search and that it uses LLM. So what does this mean for the network administrators?

Let’s unpack a few of these terms and then analyse it. To start off with, GenAI is a system that can generate content which can be text, images, etc. based on a query. For this to happen you need to have AI models so that the AI algorithms can be trained and learn. Now that’s where Large Language Models (LLMs) comes into the picture.
LLMs are systems that are designed to understand, process and analyse natural language data and then use that information to automatically generate responses to user questions. LLMs are trained on large data sets that are often measured in petabytes.
The landscape of LLMs is indeed diverse, encompassing open-source LLMs like LLaMA 2 (introduced by Meta), BLOOM to mention a few and closed-sourced LLMs like Bard, ChatGPT, etc. Lastly there are quite a few industries that are building their own private LLM as well as. To reach their full potential, LLMs are then coupled with the private data (industry specific one like in our case data networking).
Now to take advantage of GenAI based Search for Network management and operations platform, one can take the approach of

1. Making API calls to public based LLMs and then show the results
   • This is attractive if you don’t have
     • significant data lake of your own and/or infrastructure to provide its pre-processing.
     • the time or infrastructure to leverage AI safety by removing Personal/Customer Identifiable Information (PII/CII) and have a sandbox approach
2. Use One’s own specific LLM models
   • Aruba Central has taken this approach since it
     • is hosted on HPE GreenLake Cloud Platform (GLCP) which enables it to improve user experience, efficiency, and privacy by using the self-contained multiple LLM models specific to our product and access to all the resources of GLCP.
     • has a very large endpoint data lake. The last I heard was 1B, yeah One billion sanitised customer endpoints.
     • has over 4 million managed devices aka APs, LAN Switches and Gateways.
     • as a side note HPE GreenLake has been offering LLM cloud service for quite some time.

Aruba Central uses this approach to provide fast, accurate, relevant response, auto fill AI search that is protected by railguards and an opportunity to provide feedback to the AI search.

Let’s go through some examples.
When you click on the AI search in Aruba Central , you can see your recent searches and some suggestions based on your Aruba Central configuration. Like if you only have APs and not LAN switches or gateways, then the suggestions are around AP.

Also note that all the information that is displayed is based on latest HPE Aruba authoritative content and not like a classic search engines that often don’t retrieve the latest HPE Aruba content. Then as you type your search query, it tries to autofill it with relevant info, in the following example it is about Cloud Authentication.

Note that it can also use Natural Language Processing (NLP) in queries, like the one I am showing here.

The important thing is to note the accuracy of the results. And in the bottom of the search result you can send help it to improve its result further by clicking on thumbs up/down icon.

Remember that one of the features of a good GenAI search is to have guardrails that helps with reliability and security of the search function. Here is an example of one asking about spicy sauce.

Another interesting aspect of the GenAI search is to help identify issues and then assist with Root Cause Analysis (RCA). Let’s tackle wireless roaming issue.

Note the response was very specific with 5 Clients impacted. Clicking on the “View Button” takes us to

And this search example for picking up PoE issues on the LAN switches.

Again, note the accuracy of the information that was provided. Even in a small to medium size campus you could have around 60-100 switches and see the value of being able to find these types of issues so quickly.

Gen AI based insights/suggestions are only as good as the data quality it is derived from. The larger the data lakes that these GenAI learn from, is a big contributing factor to the accuracy of AI Insights.

Here is another example, imagine you are new to Aruba Central and don’t know for example, how to go to SaaS Express dashboard. You know the senior members of the network team have mentioned it, and that it is configured but you don’t want to ask them to show you. So you just search for it.

And the final example for quickly wanting to see APs in a particular site using AI Search

Note that the AP dropdown card is active providing useful info like AP name and their online status, you can then click on each of them and get the AP view of the network.

In summary, the continued growth of LLM models and their expanding applicability, coupled with the consolidation of more data sources, is expected to lead to rapid improvements in training data. This, in turn, will result in better and more accurate outcomes. The power of GenAI based search is evident, making it worth exploring.