Edge Computing or hybrid cloud?

Edge computing has been gaining much recognition and hype since I started working on it 5 years ago. I am in a fortunate position to have explored it, as an analyst, being one of the early participants of ETSI's Industry Standardization Group on Multi Access edge Computing (MEC), then develop and deploy it as an operator for Telefonica group and now, back to advising vendors and service providers on the strategies and challenges associated with its development.

One of the key challenges associated with edge computing is that pretty much every actor in the value chain (technology vendors, colocation and hosting companies, infrastructure companies, telecommunication operators, video streaming, gaming and caching services, social media and internet giants, cloud leaders...) is coming at it from a different perspective and perception of what it could (and should) do. This invariably leads to much misunderstanding, as each one is trying to understand the control points in the value chain and assert their position.

1. Technology vendors see a chance to either entrench further their position, based on proprietary, early implementation or disrupt traditional vendors oligopoly, based on open (source) disaggregated networking. Traditional blue chip vendors see also a chance to move further down the path of black box networks replacement by white box.
2. Colocation and hosting companies do not quite see why edge is that much different from cloud hosting that they have been doing all along, but are happy to jump on the bandwagon if it means better margins.
3. Infrastructure company see a chance to move up the value chain, by providing differentiated value added real estate and connectivity services.
4. Telecommunications operators tend to see edge computing as a possible opportunity to rejoin the cloud war, after having lost the last battles. The promise of futuristic 5G-like services for drones, remote surgery, autonomous cars, etc... is certainly what they have been communicating about but that is not going to materialize tangible revenue streams before 5 to 7 years. There are other short term revenues that can be created by the technology deployment.
5. Video streaming, gaming and caching services feel that they have been the edge pioneers, with specialized services or physical slices, deep in cloud and teco networks. They tend to resist the move from physical, proprietary appliances towards the open, multi-tenant model that would make the business more profitable for all.
6. Social media and internet giants ted to feel that there is something they should, or could do there, but most of their infrastructure relies either on their proprietary private cloud or on public clouds and it is unclear whether these models are compatible.
7. At last, the cloud leaders certainly see edge computing as a growth opportunity to offer differentiated cloud services and performance, but again, they are unsure whether to push the limit of their cloud or integrate with others.

I feel that we have started from the wrong foot here. There is no such thing as edge computing. There is a cloud, and there are devices and data centers. The largest, most impactful performance move cloud can make is to integrate with the last mile - the telco networks.Where you want a workload to run, a dataset to reside, a pipeline to transit through should be the result of:

• What is available in terms of capacity
• What is your budget / needs in terms of workload, performance, latency
• How much it costs / what is the price to run where
• What are the legal / regulatory restrictions with respect to locality, sovereignty, privacy...

The rest should be easily enough programmatically calculated. For this to occur, there is still much work to be done. The "plumbing", which is how to connect and administer heterogeneous clouds is almost there. The largest effort is really for these industries to come together on the reservation, consumption and fulfillment model. We might be able to live today with Amazon, Microsoft, Alibaba and Google cloud models, but we certainly won't be able to accomodate a lot more.

This means we need an industry wide effort for cloud hybridization at the business layer. It is necessary for all network operators to present the same set of APIs and connectivity services to all the cloud operators if we want to see this market move in the right direction.

A brief history of the cloud

Our industry is undergoing a sea change. Entire service categories where network operators had a leading position have shifted to new entrants.

It started with app stores and social networks for portals, then skype for voice , then WhatsApp for messaging, then Netflix for video…

Each of these disruptions have been the product of ambitious digital services strategy, coupled with technological advances that left previous generations instantly obsolete.

Economic population disparities have also been an important trend to consider. Whereas in the past, a large portion of the population wanted a robust, reliable service, we have seen market segments that used to be marginal emerge in strength where affordability has become more important than reliability.

The "best in class", no longer affordable for most, has shifted to "good enough" but free or inexpensive for many.

The new digital companies, unencumbered by analog technology legacy, started discovering new ways to scale up explosively IP infrastructure to sustain their growth.

The webscalers were born, and with them, the idea that infrastructure does not have to be dedicated to a single purpose or service but can be shared across services and geographies. This came from the observation that adding order of magnitudes of capacity per year was impractical, slow and costly with legacy networking.

Legacy networking was assembled of specialized appliances, dedicated to a single purpose, integrated into a coherent but rigid network topology. It was reliable but not efficient, stable but complex.

The first breakthrough was to use commercial of the shelf hardware. Unspecialized servers, which could make up from the lack of relative performance compared to specialized appliances with a much cheaper price point. If a server is 3x less efficient than an appliance, but 10x cheaper, it is an easy calculation. Moore's law accelerated this shift tremendously.

But, with more servers and physical elements came more operational complexity. The second breakthrough was to remove and centralize the management of each of these servers. Whereas in the past, it was necessary to physically access each server to configure it, troubleshoot it, the new method allowed to remotely access any server in any network, and furthermore to provide a consolidated, centralized view of the physical attributes and health of a whole network, from a single web portal. Beyond simple monitoring, increasingly sophisticated management routines were built in, to deploy, manage, scale programs remotely.

The ability to manage software remotely meant not only that the network manager didn’t need any longer to be physically where the data center was, but also that it was possible to decentralize and deploy several data centers in different geographies and manage them remotely. The cloud was born.