Close

HPE Aruba Networking Blogs

Wi-Fi That Works Means that It Learns From Your Peers

By Dave Fehling, Data Scientist, AIOps, Aruba

Cutting Through the AI Mist

In this blog post in the Cutting Through the AI Mist series, we’ll be putting together techniques that we’ve talked about in previous posts:  environmental features and peer clustering to provide concrete and actionable insights for our customers.

Before we hop in, we need to define one further thing: airtime efficiency. Airtime efficiency is a metric that we’ve come up with to measure the overall Wi-Fi performance for an access point (AP), a building, or even an entire customer. It’s a proprietary metric, so we can’t go into too many details in this post, but suffice to say that if an access point is providing high SNR links with high speeds on bands accessible to most of the client devices present, it will have a high airtime efficiency.

As an example of airtime efficiency, let’s take a look at two different APs – “AP A” in red and “AP B” in blue below in Figure 1. These APs have the same model, similar environments, and approximately the same number of client stations connecting each day. For a given day, we can look at a couple metrics related to airtime efficiency: client uplink SNR (signal-to-noise ratio) and client uplink speed. While not perfect metrics, they can still give us an idea of the network performance.

Figure 1. (Left) The average client uplink SNR across all clients connected for each hour of a given day. (Right) The average client uplink speed across all clients connected for each hour of a given day.

Figure 1. (Left) The average client uplink SNR across all clients connected for each hour of a given day. (Right) The average client uplink speed across all clients connected for each hour of a given day.

The two APs are fairly close in SNR, averaging 30-40 dB throughout most of the day. There are a few periods where AP B does much better than AP A, but the difference over the day isn’t as significant. However, there is a significant difference in the client uplink speeds attained by AP B and AP A, with AP B average 2-3 times greater speeds. This could be due to a hardware issue, and we have mechanisms in place to check that, but in this case the only real difference between these two APs is their configuration settings. Now let’s take a step back from APs and look at building performance.

Setting up Wi-Fi for a building involves a lot of specialized knowledge and work. The average number of client stations as well as the peak demand must be estimated. Enough APs must be present so that there are no coverage holes, but not too many such that they interfere with each other (and balloon the cost of the project.) There are many other factors that must be considered, but even so, the initial deployment is only a small part of the overall networking picture. Changing client behavior and demands can tax even the most well-designed network.

Enter one of our AIOps solutions, configuration recommendations.

There are about a dozen different knobs that can be tuned to optimize the performance of a wireless network, including transmission power and number of channels. We can take the guesswork out of optimizing a network by leveraging AI and our large dataset of Wi-Fi performance to recommend optimal configuration settings.

We calculate the airtime efficiency for all our customer networks and save it along with the environmental features and configuration settings for each building. Then, for a candidate building, we find all the similar buildings by environment and use AI to consider the possible improvement in performance gained by switching to each configuration set from the top peer buildings. If we find a significant improvement, and our AI model has high confidence in it, we suggest the new settings to our customer. Figure 2 has the result of such a process.

Figure 2. Plot of airtime efficiency vs time. The grey line is the origin. The red line is the time the recommended configuration changes were applied.

Figure 2. Plot of airtime efficiency vs time. The grey line is the origin. The red line is the time the recommended configuration changes were applied.

The airtime efficiency is plotted for one building over time. Our AIOps solution found a candidate configuration set to improve the networking performance for this building, and we suggested a change to the customer. The change was implemented at the time indicated by the red-dashed line and the change in performance was immediate. In this case, an improvement of about 50%. What’s even better is that as we and our clients improve their networks, the AI gets better options to pick from and a building can be reoptimized in the future as our understanding changes. Also, if the actual utilization of a building changes, our algorithm will detect that and suggest appropriate changes. What’s more, we’re actively working on ways to make this even easier for our customers, including a mechanism to apply the recommendations automatically.

As a final thought, let’s revisit those two APs from Figure 1. In actuality, they’re not two different APs, but rather the same AP taken on the same day of the week, a week before (AP A) and a week after (AP B) the configuration changes recommended by our algorithm shown in Figure 2. By using our AIOps solution, our customer was able to improve the speeds provided by a factor of 3 while also slightly improving the signal quality (SNR) of the connections, effects which were captured by our measurement of the building’s airtime efficiency.

Read More About Wi-Fi That Works

AI is Only a Tool. The Payoff is Automation, Network Assurance and Great User Experiences

Wi-Fi That Works Keeps Up with Your Business

Wi-Fi That Works Has Zero Trust Built in from the Edge to the Cloud

Toward AI-Operated Self-Optimizing Networks

How AIOps Leverages Network Environment Factors to Deliver Reliable Automation

Aruba AIOps solutions for IT