The Overlay Decade

Rewind to a decade ago, and it was common to hear the complaint that networks know nothing about the application layer, and what happens at the application layer is more important than what happens at the network layer. While that framing is worth challenging, there is an essential truth, network infrastructure is a means to an end, it is not the end in itself.

The early Internet innovators understood this, which is why something called TCP/IP was created. There was a clear understanding that the end-to-end experience was coupled to what happens in the network. Those companies that do not get this in the depths of their soul, do not understand networking in any meaningful way. Even today, where TCP and IP are in some sense separate domains, many IP routing professionals, have a good foundation in TCP basics.

Image Source: PxFuel

Over the last decade, where the importance of application performance came to the forefront, Palo Alto Networks disrupted the “network” security market by providing visibility at the application layer, something that was labelled “Next generation firewalls”. SD-WAN has also emerged, with the ultimate goal of (re)routing traffic based on Enterprise policy, including application layer performance.

The critical property of SD-WAN is that it is an approach to application performance, over a network, that is not restricted to the footprint of a single network/network operator. ANY solution that attempts to provide application performance on only a single network/network operator, is missing the end-to-end reality of Enterprise networking, where an Enterprise may use many different access modes, sending traffic to many different clouds, not just a corporate data center. Theoretically, an application performance paradigm limited to one network might have value for applications/services hosted by that network, but that is a limited value. Apples and oranges should not be mixed.

Most SD-WAN approaches use tunnels that operate above the IP layer and are not sourced/sinked in the network itself. Juniper’s 128 Technology acquisition has recently drawn attention to an approach to session-level results without tunnels. Any way you cut it though, the goal of SD-WAN is an end-to-end goal, not a single network goal. The world has changed. Leading Service Providers get the threat and opportunity of SD-WAN. The threat to their current MPLS VPN business, and an opportunity similar to what wireless gave them, the ability to expand beyond their wired footprint, however, unlike wireless, anywhere in the world. This is part of the large march of history in telecom, transforming it from a closed ecosystem that developed special standards for its own domestic market, to a truly international ecosystem, leveraging the scale and reach of Enterprise IT.

Enterprise IT is not always a better fit for Service Providers, but if the Service Provider tech is not 2-10 times more compelling than the Enterprise tech, then the Enterprise tech will take over, in the due course of time. IP and Ethernet are just a few examples. SD-WAN is the next example currently on the march. Enterprise tech serves the world. SP tech serves SPs. Enterprise tech has larger scale.

I really enjoyed the Juniper/Mist webinar today where the Mist CEO/Founder Sujai Hajela and Product Manager Joanna Chavez talking about the user experience and real AI. I still stand by my previous comment. AI is both overhyped and underhyped. Those doing automation and AI washing are noise, with little signal, and are getting left behind. Ditto for those who don’t get it about how this new tech will change networking. Those developing real AI, are moving the needle forward, and in so doing, moving themselves into a leadership position. Importantly, not replacing operations people, but augmenting them.

Yes, it is about the user experience. Yes, the difference between automation and autonomy is whether you do the same thing twice or not. The mission at Mist is no repetition; the mission is learning, and working out how to get to a better, autonomous result. By complete uncoordinated coincidence, this is similar to the definition I laid out in a recent article on Augmented Routing.

Repeatable workflows are important, but there are any number of non-equipment companies that will/do play in that space. Autonomous networking is where networking suppliers can and should move the needle. Networking experts working with learning machines to totally change the experience of network operations and end-to-end experience.

When I finish my Network 2025 update, I will write more about the underlay vs the overlay, but in short, overlay tech is likely to progress/accelerate faster than underlay tech. There are a number of reasons for my conclusion on this, including present day extrapolation, but in short, they relate to: the ability to experiment, size of the failure domain, and complexity of problem.

If not already obvious, the last decade is characterized by the emergence of the overlay, whether in the WAN or in the datacenter (server-to-server virtual networking, for example). Finishing and refining the overlay work that has been started, will be an important industry focus between now and Network2025. Today, the terms overlay and underlay are commonly used parts of the networking vernacular, and I do not recall that being the case a decade ago. IP/Ethernet have pushed down into the transport layer replacing SONET/SDH, and limiting the applicability of OTN. As this has happened, overlay innovation has also emerged.