Mist has announced enhancements to its WLAN offering, including improving anomaly detection with newly added algorithms to analyze performance and system data, boosting the natural language processing capabilities of its Marvis virtual assistant, and automating Radio Resource Management (RRM).
I’ll walk through the key announcements, but here’s a little background on Mist for context first.
Mist is a startup that competes with the likes of Cisco, Aruba Networks, and Aerohive. It sells cloud-managed wireless APs for enterprise WLANs, as well as location-based services and inventory tracking.
The company positions itself as an AI-driven IT infrastructure company. It says it collects more than 100 data points on APs and wireless clients every two seconds. It forwards these data points to its cloud service to feed its machine-learning algorithms.
The company claims this data collection and analysis enables it to provide near-real-time metrics on Wi-Fi performance and user experience, set and monitor service levels, and provide useful recommendations for troubleshooting when problems occur.
While Mist provides a dashboard for WLAN and network administrators with charts and graphs, it prioritizes its virtual assistant, called Marvis, as the mechanism through which administrators will interact with the system.
In particular, Mist says administrators can type natural language queries into its interface. This natural language feature, which was announced in February 2018, has now been enhanced to understand hundreds of colloquial phrases and queries, up from dozens when it was launched.
For example, an admin could type “How’s the AP in conference room 3?” or “What went wrong with Jane Doe’s iPhone yesterday?” and then get a response that may include multiple answers, as well as pointers to potentially related issues.
The goal is to alleviate the need for administrators to get answers without having to search through multiple dashboards or metrics.
Mist also says it has improved anomaly detection by incorporating new algorithms into its machine learning system to parse the metrics it collects to establish operational baselines and then look for deviations. When an anomaly occurs, Marvis can alert administrators.
The company also offers APIs so that anomaly alerts can be integrated with ticketing or help desk systems to kick off workflows to investigate and remediate problems.
Finally, Mist is announcing automated Radio Resource Management. Mist APs are continuously measuring RF information on coverage, capacity, channel usage, transmit power and other metrics.
For customers who are comfortable enabling automated changes, Mist says it can adjust its APs based on real-time conditions. It then measures the effects of any changes to ensure user experience stays aligned with administrator-defined performance metrics.
Don’t Trust, Just Verify
I’m skeptical of anything that’s served up with healthy portions of AI or machine learning. That’s partly because I haven’t learned enough yet about data science to discern what’s actually feasible in an IT product and what is, let’s say, ‘aspirational.’
It’s also because, in my experience, there’s usually a gap between a company’s marketing message and how the product actually performs in the real world.
Take anomaly detection. Based on my understanding, anomaly detection is a good use case for machine learning (assuming you have a data set of sufficient size and the right training and classification systems in place). Machines are good at taking in large quantities of data and spotting differences.
The question then becomes, how useful is it to be alerted about that anomaly? What’s the context around it? Does it signify a larger problem, or is it really just a blip that won’t surface again? Is there a mechanism to provide feedback to the machine learning system so that it better understands what’s relevant and what isn’t?
In other words, the trick for customers (and tech bloggers) is to figure out not just whether a product actually has the capabilities it says it does, but also how well it executes on those capabilities.
This isn’t to cast aspersions on Mist. It’s entirely possible the company’s products do make good use of machine learning algorithms to make a WLAN admin’s job easier.
I just want to make it clear to anyone reading this that I don’t have a strong sense of the size of the gap between what Mist is telling me, and what it can actually do.
Therefore, if Mist sounds like a fit for you, make sure you try before you buy, and see if you can get honest feedback from existing customers.