A Paradigm Shift in AIOps Networking

A typical network may experience many sources of issues and anomalies. From external issues such as DFS Events or interference, internal network issues such as insufficient power, bad cabling, or issues with clients such as poor software or sticky clients, the reasons are multifold. Ultimately, we want to do one thing - optimize the user experience for those using our networks, without spending hours tweaking our network every day. This is why we released ExtremeCloud IQ CoPilot last year, which helps customers to proactively identify anomalous network behavior (or validate that it’s not me), and to know and to fix those anomalies. Furthermore, CoPilot is extending its scope to digital modeling with a recently announced capability (discussed later).

When it comes to pesky network issues, IT Administrators spend most of the time trying to Identify and Know the reasons for the problem. Once a problem is spotted and the root cause becomes known, resolving the problem is typically trivial. While CoPilot helps proactively prevent problems in the first place, it also seeks to drastically reduce the overall Mean-Time-To-Resolution (MTTR), which generally consists of Mean-Time-To-Identify, Mean-Time-To-Know and Mean-Time-To-Fix the issue. Let’s look at each of these categories and the associated tools within CoPilot that help contribute reducing the time spent on each of these stages.

Mean-Time-To-Identify:

The Events and Alarms capability allows each network to have custom alerts based on pre-defined sensitivity levels. ExtremeCloud IQ Companion is a mobile app that allows IT to get alerts about issues that happen in the network, even while they are on the go or away from the office.

Mean-Time-To-Know:

Instead of searching a needle in a haystack, what if you got a simple view of how wired and wireless clients experience the network? What if you got an ability to look through the various anomalies in your network and the data that led to this conclusion? The following screenshot shows the Connectivity Experience for wired/wireless users, giving you the intuitive context and insights to help reduce the time you need to know what is happening to them.

Mean-Time-To-Fix:

Explainable Machine Learning algorithms are a subset of the full spectrum of ML algorithms that allow you to understand why a recommended fix is suggested. For help with fixing issues, CoPilot provides videos and detailed knowledge base articles tied to each anomaly that are linked from within ExtremeCloud IQ. Finally, if IT operations teams are unable to remediate the issue themselves, then a GTAC case can be opened within the app itself with all the associated information, helping to simplify IT operation workflows.

CoPilot can help to significantly reduce the MTTR for IT teams, but the lifecycle of network issue can often start much sooner. A lack of understanding of network devices and their capabilities can doom networks long before deployment. Due to this lack of understanding, organizations frequently invest in test lab space to understand how their devices will operate in the field. While testing can help organizations validate their network devices, it can also add CapEx and OpEx to their bottom line. In addition, the pandemic has thrown significant wrench into testing labs. IT teams are working remotely (read – who is the lucky person to physically truck-roll the equipment to the office lab and rackmount the equipment) and supply-chains are delaying the procurement of physical equipment (can I get a switch pretty please?). At the same time, networks are more mission-critical than ever. So, the cowboy attitude of rolling changes out on Friday nights and working all weekend to get those patched before Monday morning is no longer the preferred modus operandi.

Enter Digital Twin

In the last decade, cloud offerings are increasingly preferred over corresponding on-premises ones. The growth of advanced cloud capabilities allows us to process data in a whole new light – creating a digital model of a network device. This concept is not a new one. From NASA creating models of a MARS-rover, Boeing making a digital model of their airplanes, to cities making a model of their infrastructure and corresponding ecosystem. Many sci-fi movies follow the ”modeling” theme - from the Amazing Spiderman that showed a lizard being modeled re-growing its limbs to the iconic Attack-of-the-clones. So why are computer networks left out?

First, let’s define what a digital model is - a digital model is a virtual replica of a physical object. So, what if you could create a model of a switch, select your operating system on the fly, along with an optional policy that you may want to apply and hit ‘Launch’. You can now observe the switch user interface within ExtremeCloud IQ or open a command-line-interface (real switch engineers love CLI!). Viola, play with a digital model of your favorite Extreme Networks Universal Switch without getting off your sofa or paying for expensive test lab space! You can now familiarize yourself with the platform, OS version and configuration before having the physical switch in-hand! What would a modeled network look like?

The conversations with our customers and partners show us that the potential for this technology can be transformational in nature. From transforming how test labs are managed to training services to professional services there are many uses of digital models.

This blog was originally authored by Jeevan Patil, Senior Director, Product Management