Edge computing today may be most simplistically defined as putting an extra layer of computing in the network between your smartphone and a centralized data center in the cloud for some optimizing purpose.
From this original, slightly narrow vision, there is still some ways to go before realizing full stack relocations (e.g., relocation of an entire web server) that will allow true service enablement just one hop away from the end user device. Obvious benefits of realizing this vision include reduced latency and backhaul capacity reduction, but I believe there will be more profound benefits to operators and service providers in terms of new business model enablement, too.
Software-ization will be the key enabler
Ubiquitous software-ization of the overall networking stack is the key to real edge computing. Software-ization will enable the envisioned “full stack relocation” through virtualization technologies such as NFV and platforms such as OpenStack. More so, the inevitable software-ization of the underlying transport networks will bring about dramatic change in the networking fabric that can be leveraged for flexible and early termination of services near the end user.
Specifically, the control and user plane separation enabled by software-defined networking (SDN) technologies provides the basis for novel services being implemented as simple software extensions on top of commodity SDN hardware. This world will decouple network service providers from vendors of equipment, very much like decoupling application providers today from those selling a smartphone or laptop.
Opportunity emerging for operators and service providers
The software-ization of the networking fabric in combination with NFV-managed computing infrastructure also creates new opportunities for operators by virtualizing key parts of the full “network stack.” While today, managed computing racks for major service providers such as Google, Akamai or the BBC need physical installation in key locations of the operator network as so-called points of presence, the increasing flexibility of the networking fabric will present an opportunity to move towards a model where operators can provide computing-as-a-service capabilities for any service provider, not just the major ones (Netflix, HBO, etc.).
Key to this is for operators to utilize their footprint in managed site infrastructures by installing and renting out computing capabilities based on common (NFV) platforms that run on COTS hardware near end users in, for example, eNodeB, Broadband Gateway or Customer Premise Equipment. With this capability, operators can expose an edge surrogacy service to anyone who provides HTTP-level services today, while meeting 5G KPI such as low latency or increased throughput. With that, any available and future internet service will effectively have been moved to as little as just one hop away.
The challenges in realizing full stack relocation
Relocating a full internet stack is challenging. First, IP connections need to be terminated flexibly and in places where current network architectures do not foresee such termination, such as within eNodeBs. This requires new approaches to routing, as well as mobility handling, since current anchor-based indirection approaches will not suffice in a world of highly mobile services and devices.
But relocating an entire stack needs to move beyond just IP. Particularly in the mobile network world, this also requires a rethinking of the current bearer-based connectivity model, which effectively establishes tunnels from one mobile network component to others. This highly inflexible approach causes overheads and delays. Beyond such core connectivity improvements, current HTTP-based web services will need to redirect requests to the nearest service endpoint, possibly just one hop away or located in a nearby mini data center.
Currently utilized DNS indirections do not meet the flexibility or the timing requirements required for real edge computing, where decisions on selecting service endpoints might be based on near real-time criteria such as network or server load. Solutions to those challenges currently being incubated in research and standardization communities must eventually integrate with the SDN-enabled transport networks that will ultimately proliferate in 5G mobile and fixed networks, while not overburdening the control and management for these solutions. This in itself calls for more innovation in self-management of networks, particularly at the edge.
A roadmap to the real edge: 3 steps
I see three steps as being key in the roadmap to realizing the real edge.
First, the discussions in the relevant forums need to evolve towards the full stack relocation vision that I outlined. We need to move beyond the small-step evolution that is embedded into today’s network infrastructure vision where the access merely assists an end-to-end service from the user’s device to a remote data center.
Instead, the access network has to become a full member of the end-to-end view of the service with the possibility to fully terminate, as well as place a service as close as one hop away from the user.
Second, core enabling concepts need to be developed as concrete solutions that can be tested and pushed into standards. This early stage research, such as the EU-funded Horizon 2020 program, is happening right now in international 5G efforts, publicly as well as privately funded. It has led to key insights in the areas of SDN, NFV, flexible routing and service surrogacy, all being key enablers to real edge computing.
These early stage research efforts need to continue and graduate into early deployments and test beds. I am seeing this vision of edge computing coming to life within these communities.
Last but not least, this acceleration of research efforts needs to be aligned with necessary contributions to standards such as ETSI Mobile Edge Computing Phase 2 (due to start mid 2017), IETF and 3GPP, where key foundations for service indirection, control plane design and other important aspects for edge computing will be formalized.
In addition to standard bodies, governments are waking up to the readiness, as well as openness, of their national infrastructure that would enable novel and crucial services for a plethora of societal challenges. Real edge computing will play an important (perhaps not yet fully understood) role in this discussion regarding the future of the internet.