Making Edge a Reality

February 19th, 2019

Headshot for Jason Hoffman

Jason Hoffman

President and Chief Executive Officer

We believe that the overall value of edge is to create and be additive to a value chain by connecting devices and clouds in a way that is end user, application and mobile network aware.

We give an overview of our architecture and how the parts satisfy needed missing pieces in the industry: how to aggregate existing infrastructure assets into a common market, how to allow for trading and clearing between everyone and how to expose mobile infrastructure to applications on both devices and clouds and complete the chain.

We advocate that the superior developer experience is the “pull model” where the appearance of devices and applications trigger a streaming lambda deployment that pulls software into the edge infrastructure from clouds and then works in contextual concert with mobile networks and their connected devices. We contrast this to the inferior “push model” which we state is inferior because it is not aware of mobile networks or end user needs in a specific geography and at determinable times.

And finally, we state what we believe to be a unique and complete viewpoint that has allowed us to have a long term roadmap around a portfolio of edge native services that are common in present and future use cases. We detail a framework that shows that AR/MR, Autonomous Vehicles, Blockchain, Computer Vision and Machine Learning/Artificial Intelligence are in fact just technologies that allow for an evolution of content delivery. This is the basis of the framework that we use to connect all the disparate use cases to developers’ needs and to the next decade of mobile network evolution.

How do we see the overall value chain?

By the time that you read this, MobiledgeX has been operational for 414 days. With the release that we’re announcing, I believe it’s appropriate to reflect and take this opportunity to clearly explain our viewpoint.

We were founded by Deutsche Telekom to determine and execute on a path forward for “edge”. Deutsche Telekom “is a German telecommunications company headquartered in Bonn and by revenue the largest telecommunications provider in Europe.” In the US, we’re familiar with them as the majority owners of T-Mobile and the progenitor of the prominent magenta “T”.

Mobile networks are commonly experienced through the use of our smartphones. Our smartphones are the most personal of personal computers, a device trusted by our banks, and their use has been the basis of tremendous disruption and value creation over the last twelve years (and yes, I’m only starting the clock with the launch of the first iPhone: my apologies to Handspring, Palm, Blackberry, Nokia and Ericsson). Smartphones connect to base stations in the Radio Access Networks, and globally there have been millions of base stations installed. By all measures, mobile networks are the largest distributed, technical and economic machine deployed on the planet. Smartphones and base stations are connected by specific frequency bands (“spectrum”) of radio, and this spectrum is a government-controlled asset class in every country that has a government.

Along with smartphones, hyperscale clouds have emerged. While several network operators attempted efforts in “cloud”, these efforts were typically done within an existing enterprise hosting business (if one happened to be present) and were not part of the mainstream investments in the mobile networks. With the trends occurring now, this has changed. The lessons from hyperscale clouds are increasingly incorporated into how mobile networks are designed and operated, and the network functions that form the network itself have been (and are being) redesigned to be yet another service deployed on cloud infrastructure.

We can think of the basics: smartphones connect to mobile networks, and mobile networks connect to clouds. To provide some numbers, I’d like to use Deutsche Telekom and the typical public cloud in Germany as an example. There are approximately 60 million smartphones in Germany, a subset of which are subscribers. These connect to base stations on ~40,000 towers, the towers connect into 900 access sites, which connect to 11 mobile core network locations. The mobile core connects into transport networks that connect to 3 Availability Zones present within Frankfurt.

An application deployed on a cloud’s 3 availability zones in Germany will send out data that then goes through one of the 11 sites, then through one of the 900 sites, out to 1 of the ~40,000 towers and then finally to one of the 60 million smartphones. Smartphones are even allowed to travel in our cars at high speeds on the autobahn, hopefully maintaining seamless connectivity while switching towers and access sites.

For us, ”the edge” is a business that we can do by actually participating in this value chain, not just connecting clouds to devices. It is not about edges replacing clouds, it’s about the global operator community stepping up to allow the safe use of the infrastructure that they’re modernizing and building anyways to make one dynamic, distributed system that is easy to consume by application developers.

What then are the tasks and how do we do them within a software system?

The first task is to have a common, realistic understanding of the infrastructures that are available to us along with parameters beyond the technical components: scarcity, time, unit economics and more. This is the part of the architecture called the “cloudlet resource manager”.

The second is to have a market understanding and to allow for trading. This part of the architecture is the “distributed matching engine”; it is not a coincidence that distributed matching engines are common components in high frequency trading and clearinghouse systems.

The third is to have a controller structure that coordinates with both of these and exposes interfaces on both sides of the infrastructure: one towards the devices and one towards clouds.

What are the best developer experiences that we should create on each side of the mobile network infrastructure?

This is where we have a unique viewpoint about the consumption experience. There are two logical possibilities or perhaps you could call them philosophies.

The first is the “push model”, where there’s an expert, imperative, programmatic interface that allows cloud operation teams to push their applications into the edge infrastructure. We do not believe that this is the easier experience: it fails to abstract a more distributed system, we leave the “how” up to those who deploy and operate the applications, not those that develop them and there isn’t the opportunity to use additional edge native services that are present.

The second is the “pull model”, where there is an application- and device-triggered pull from cloud into the edge. This is the model that we believe is the correct one. A device shows up on the mobile network, an application is launched and its presence triggers a git to cloud to edge streaming deployment where we can automatically take care of the “how” the application is placed, scaled and life cycled. For this, we have native device integrations and SDKs in all the major languages. Application developers should have straightforward additional “if edge, then awesome” APIs with which they can develop against in their existing environments.

And now to the final question: what’s a straightforward way to think about how the future will unfold?

We’ve said that the trend in the mobile operator networks is an industrialization of the underlying infrastructure in a hyperscale cloud model and then the disaggregation and further distribution of the network functions. As these transformations are occurring, there will be greater capacity and capabilities present within the mobile infrastructure (the stuff that’s never going away because it has to exist to connect clouds to devices). We can think of latency as a continuum: there will be 30 down to 20 milliseconds, then 20 down to 10 milliseconds, then 10 down to 5 milliseconds and 5 down to 1 millisecond windows that will open up over time as we modernize into 5G networks. Each latency window will make possible a different set of use cases, additional devices besides smartphones and new opportunities to create tremendous value for everyone.

The question though is whether there’s a common set of functionality and features that should exist as edge native services throughout these improvements that can be shared amongst use cases?

We believe there are.

Let me anchor my explanation of this in the most common thing we see in mobile networks today: media delivery (aka content delivery). We have all watched movies on our smartphones: the movie is streamed down, takes over the entire screen of the device, does not have any input from other components, and anyone who views it watches the same movie.

I believe that there are three future phases to “media delivery”.

The first is where a single character is streamed to a device and has to be mapped in a live feed from the device’s camera. The digital character (for example, a Pokemon) also exists in a specific location in the world; if you move the phone, it realistically remaps on the background; it has individual behavior personalized for you, the end viewer; you may control them or they may be autonomous; they could be “owned” by you or different users/companies. Now add multiple characters. Then add multiple viewers with different smartphones. Then add multiple mobile networks. Then expect them all to be a high fidelity and collaborative, social or multiplayer experience. These are the challenges of AR/MR, broken down. Enabling each of these is non-trivial and requires specific edge services to be present along with the application back-ends within the mobile network infrastructure working together with a presence in clouds and on-device.

For the second, let’s imagine we change from delivering digital content to delivering physical content with the same characteristics: the “content” will use cameras to map and understand their position in the world; they will exist in specific locations; they will have individual, personalized behaviors; they have to collaborate in a social setting; they may have a human controlling aspects or they are autonomous; they could be “owned” by you or different users/companies. Examples of physical content are cars, trucks, drones, robots. As you can perhaps guess now, we believe that the entire ambitions around autonomous and connected everything require the same edge native services as AR/MR; however, they also require future latencies and capacities, while multi-player, mobile AR games are possible on today’s networks.

And for the third and final phase, media delivery will start flowing in the opposite direction. Today it is delivered out of mobile networks, and as we’re putting computer eyes on everything, there will need to be capabilities that take in, view, identify, learn and act on that incoming video. In other words, large scale video ingest along with computer vision and AI services will come into their own. This is something that we will have to do considerable work to enable and why I’ve placed it last.

To the final question.

Can we engage now?

For the first time, 3rd parties are deploying directly to private mobile operator edge infrastructure without any human involvement, dynamically allocating their workloads as close to their mobile users as possible. We call this “git to edge”: the automatic allocation and deallocation of application backends as close to end users as possible, finally solving full application mobility.

Bringing the cloud to the user is a pre-requisite for the new pervasive and immersive experiences that require a real-time application of machine intelligence and spatial understanding combined with low latency and high scale.

MobiledgeX is already working with SK Telecom in South Korea and Vapor, the Kinetic Edge Alliance, in the USA, and more operators are soon to be announced to create one seamless global mobile edge and release the capabilities of existing global infrastructure and data, both for operators to monetize and application developers, device makers to benefit with proximity and dynamic execution.