Mist – Applying AI to perennial WiFi connectivity problems

While at Faultline Online Reporter we have assiduously followed in-home WiFi and hotspots, we have perhaps neglected the application of WiFi in the enterprise. When we last looked (ten years ago) we continued to see enterprise WiFi built around on campus controllers which cost extra and had to be tuned.

But a 3 year old start up called Mist claims to be changing all of that, by moving far more of the logic to the cloud, using open source big data tools to collect and shift millions of data points and machine learning apps in the cloud which can then retune an enterprise or campus WiFi environment automagically.

We talked to Jeff Aaron, VP Marketing at Mist, just as they opened for business in Europe last week, having landed some key accounts in the 18 months since they opened for business in the US, including Amazon and Facebook and Mercedes Benz.

The company has two main produce sets, better WiFi, but also Bluetooth Low Energy (BLE) for IoT applications, which claims to eliminate the complexity in beacons and offer location based apps, which for instance drive welcome messages at retailers, health departments and hospitality suites. This was a bit off subject for us, but it uses a patented dynamic BLE 16 antenna array and when combined with machine learning in the cloud, gives 1 meter location accuracy with sub-second latency, using what it calls virtual beacons, which are calibrated automatically in real time by machine learning.

Aaron gave one retail example in the US which we took to be Petsmart because it was a pet shop with 1,500 stores, which chose both systems – BLE and WiFi, to fit out its entire US operation, where the WiFi had to enforce key metrics like time to connect, throughput and roaming and support Voice over Wi-Fi, allow rapid guest access and always offer employees capacity for hand scanners and printers.

Mist went up against Aerohive, Meraki and Aruba to win this particular retailer deal, and consistently comes up against them. It now says it can use the Cisco APs instead of its own, if any customer wants, with a layer of its own software running on them. “We don’t want to be in the hardware AP business,” Aaron insists.

Today the APs have to be bought from Mist, because it requires one or two adjustments to the Broadcom 802.11ac chips to accommodate the transfer of metadata into the cloud.

“We re-wrote the control plane and can completely control the user experience, for instance guaranteeing sub 2 second connect time, or 10 Mbps at each device. And if this is physically impossible to achieve, the AI in the cloud can recreate the problem and work out a strategy that will likely fix it and makes recommendations to the APs to change in real time. It even offers the support team natural language processing to talk the virtual network assistant. It can even do prediction on a particular workload and tell you what you need to change in a configuration if it is not possible with the current one.

This is somewhat analogous to in home multi AP WiFi, which uses client steering and band steering to get as close to the theoretical speed maximums by sharing load balancing workloads from different mobile clients, dynamically. These at present don’t use machine learning, and Aaron seemed unfamiliar with them, but then again they are only now converging – it used to be that in-home WiFi was on its own, one AP against the world. But with Multi-AP configurations, MU-MIMO and reporting to the cloud, it has moved to almost precisely the same spot where enterprise WiFi has reached.

Until Cisco came out with its Meraki system, it used a local controller system to manage load balancing and roaming, but with improved broadband speeds this can now be shifted to the cloud, so enterprise, like consumer WiFi is just multiple APs locally and a cloud controlling app – we have come full circle and the two disciplines may now learn from one another once again. “AI used like this has a potential applicability in the home space,” Aaron confirmed, and said the company was now working with Verizon, but mostly on enterprise WiFi.

Mist has been shipping for 18 months and has 300 customers, 30 of which are in the Fortune 500 and is looking at Europe for WiFi, applications for its virtual network assistant, BLE engagement and asset tracking and Aaron says there are lots more use cases where AI can work with WiFi.