I am a believer that the Device Virtualization vision will manifest fully one day, but we have a long road to travel to this end game. The related notion of Service Endpoint Agent (SEA) devices that I have written about is simply a continuation of emerging edge and fog computing trends that will ultimately drive the transformation of wireless networks into flexible, new, value-added service offering environments where said services are literally just one hop away. However, if we want to realize this vision, it will require that the edge transforms into a different type of mini-internet that behaves just a little differently than the wider internet that it connects to.
So what is this idea of a mini-internet
Today, there is too often a tendency in marketing literature to show the edge as a simple “box” that is colocated with some form of radio access units. The reality is obviously slightly more complicated than this. These so-called edge boxes actually form part of a plurality of Layer 2 transport networks that typically connect said radio units to a point of interconnect with the Layer 3 internet. The problem with this basic architecture in the context of the larger service ambitions of 5G is that it is inherently weak when it comes to minimizing latency. Stated quite simply, all the services we know and love in the internet exist, not unsurprisingly, in the internet, i.e., well beyond this Layer 2 transport network where all these edge boxes reside. Protocols like HTTP and DNS that provide access to these services only terminate in the internet. Therefore, in order to access a “service” now or in the future at an edge box, it is necessary to first go out to the internet and then come back in to the edge. This so-called “tromboning effect” is inherently inefficient and not amenable to the low latency ambitions of 5G.
Intuitively, the answer to this problem is to localize communication or service routing, but this is easier said than done. HTTP, DNS and other internet protocols only work at Layer 3 today. In order to break this mold and enable fast local communication, we are going to have to reinvent how this all works at the “edge.” I see this reinvention as somewhat inevitable and will ultimately render us to a multi-verse of mini-internets all working in harmony with the larger internet that they connect to.
Why this is quite hard to do but not impossible
Turning the edge of the network into a mini-internet is a hard problem because, from the real internet perspective, the edge is just a “link.” Transforming this link into a flexible service environment that accommodates inherent localization and mobility requirements requires us to fundamentally rethink components of the internet. But, what if we just push IP networking all the way down to edge – would that not solve the problem? We actually played with this idea back in the 3G days, but the so-called notion of an All IP network died on the vine quite quickly. For practical reasons, edge networks need to be highly optimized managed Layer 2 transport networks. Turning these transport networks into fully fledged IP network extensions to the internet is not really an option. What we need is a hybrid solution of some sorts that supports the envisioned new service environment capabilities but also supports legacy Layer 2 Transport network requirements.
Service routing over Layer 2 networks: A feasible solution?
The first step to moving towards a mini-internet at the edge is recognizing that service routing over Layer 2 networks is not a problem that needs to be solved by the blunt instrument of introducing a full Layer 3 stack at the edge. The key to this is the proliferation of SDN and NFV as core technologies for 5G. Both technologies together enable purely software driven innovation at the edge with a flexibility through virtualization and programmability of computing and communication resources. One such application of these technologies here is the enablement of service routing over Layer 2 protocols. Mapping protocols like HTTP and DNS onto new and more efficient Layer 2 protocols enabled as an SDN/NFV application opens up the possibilities to reinvent how these and other protocols work at the edge. The facility to quickly direct service requests to the most appropriate (virtualized) service instance in either the mini-internet or the real internet represents a transformative capability that just might be the difference between enabling a new service environment in 5G or not.
When are we likely to see this capability?
Forums like 3GPP recognized the need for this type of capability in recent work on Service-based Architecture (SBA) for 5G control planes, foreseeing the realization of control plane functions as rich and flexibly composed web services without requiring full IP networking to exist everywhere in the cellular domain. Questions are also being asked on the suitability of the DNS to quickly indirect service requests to new instances in cases of fast reconfiguration of the 5G control plane. These discussions are aligned with ETSI MEC Phase 2 work that poses new requirements on service request routing with constraints of a few tens of milliseconds to enable rich multi-access services to be realized in the edge service environment. In addition, early work on pure service routing over Layer 2 is underway in the IETF to push these ideas into its future standards. If history repeats as usual, these standards will be picked up in a not too distant future release of 3GPP and the multiverse of mini-internets will emerge.