It’s time to go back to the future. No, we’re not talking about the movie — although there are intelligent shoes involved — we’re talking about returning to a distributed model of computing. We’re talking about edge computing.
How we process and store data input has followed a cyclical trend: we started with mainframes (centralized), then moved to the client server model (distributed), then transitioned into cloud computing (centralized), and are now poised to return back to a distributed model with edge computing. To put it simply, edge computing is a system wherein devices process data at the edge of the network rather than sending data to the cloud to be processed. Its primary use cases involve IoT — including self-driving cars, wearables, and drones. And while it’s not here to destroy the cloud, it is here to subsidize the cloud with real-time data processing.
Is IoT Here to Destroy the Cloud?
The coming of age of the Internet of Things (IoT) will have a major effect on the ability of the cloud to hold up as a viable means of data processing. IoT devices have hundreds of sensors, and the proliferation of these sensors — in our shoes, in doors, in locks, in cars, and in everyday objects such as coffee machines — will result in thousands of data points. A team of 51 senior NewtonX automotive experts estimated that a self-driving car collects and processes 5TB of data per day and sends 50GB per day to the cloud. This team agreed unanimously that as this data increases exponentially, the cloud will become encumbered by two limitations:
- The speed of communicating large datasets becomes a bottleneck. Today, sending 5 terabytes of data through the cloud with a 100MB connection still takes 4.6 days without encryption and 16 hours — which is literally slower than just shipping a hard drive by “snail mail”.
- The lag in communication hinders the real-time computation required for IoT and other processes. A back and forth for even the simplest of calculations in a data center can take anywhere from 50ms to 100ms, which is dramatically slower than what is required for certain workloads such as AR/VR, self-driving cars, etc.
This poses a problem for a world that will soon be built around real-time processing of very large datasets. Take a self-driving car, for instance. When the car sees a stop sign, it will need to process that information instantaneously in order to act upon it. Similarly, a shoe would give you information on your running in real time, rather than just giving you a report at the end of the run. In a sensor-driven world, the latency inherent to the cloud simply doesn’t make sense.
This shift has already begun. Bragi, a startup in Munich, recently developed wireless earbuds that keep data on their wearers’ vital signs local. And the shift isn’t limited to newcomers: Amazon Web Services now includes a software called Greengrass, which can convert IoT clusters into micro clouds for rapid data processing. According to a at 3 senior NewtonX experts familiar with the industry, it is likely that Amazon acquired Whole Foods in order to stake out territory for local data centers (this was reportedly a major consideration in the acquisition process). There is also talk of edge computing being applied to telecom carriers. VaporIO, for instance, is attempting to “build the world’s largest network of distributed edge data centers by placing thousands of Vapor Chambers at the base of cell towers and directly cross-connecting them to the wireless networks.” Initiatives like these are sure to crop up in ever-increasing numbers as the sheer volume of data that our devices produce increases.
Looking at a Massive Addressable Market
With each new stage of computing the total addressable market has increased — and edge computing will be no exception. Every person has maximum two desktop/laptop computers (a work computer and a home computer) and that’s maximum number of computers that would ever really be sold. But when we’re entering a market where every person could have fifty different devices (counting doorbells, coffee machines, and wearables), the addressable market grows enormously.
Edge computing will grow alongside the proliferation of IoT devices. IoT devices will in turn become more and more sophisticated as they learn through larger and larger sets of training data. Currently, only roughly about half of the data that sensors collect is actually being analyzed — and as the sheer volume of data produced proliferates alongside device adoption, edge computing will become all but necessary. Along with increased usage, however, will come increased need for ever more sophisticated uses of IoT — and the sheer volume of data produced by sensors will not realistically be trainable at the edge.
Which brings us to the future of the cloud. The cloud isn’t going anywhere — particularly considering that edge computing still comes with major security concerns. Rather, we will see hybrid edge and cloud solutions, wherein data is curated and processed at the edge, and then sent to the cloud for storage and — in the case of IoT — training. SaaS applications will continue to function in the cloud, while mission-critical business processes may be computed at the edge. The cloud will continue to grow, and will be critical to the success of IoT as a hub for machine learning.
This is all to say that even in a future with millions of sensors, a future where your shoes collect hundreds of data points every time you go for a run, IoT will only succeed through a marriage of edge computing, cloud storage, and machine learning in the cloud.
The Immediate Future Isn’t Edge Computing; It’s Micro-Clouds
The world of trillions of sensors on everyday objects is not the world of tomorrow, but a much more distant future. The immediate solution to the need for rapid processing will be micro-clouds — such as the previously mentioned AWS Greengrass. These micro-clouds will still be operated by the mainframe cloud provider, but will allow for near real-time data processing. They will also be significantly more secure than device edge computing, which currently is highly vulnerable to attacks.
According to a panel of NewtonX SVP-level IoT experts, security is the number one obstacle to full edge computing in an IoT context, even before performance limitations. Micro-clouds, however, come with the same centralized security that the current cloud system has. Micro-cloud providers can team up with telecom companies to use phone towers as small data centers (which is in essence what VaporIO is doing), and essentially replicate the benefits of edge computing for most use cases other than, for instance, self-driving cars.
As one NewtonX veteran industry consultant rightly pointed out, securing trillions of devices that are operating without the knowledge of a centralized system will be a massive challenge for the future of edge. Until this challenge is met with a viable solution, edge computing will likely be handled by the same big players — AWS, Microsoft, IBM — who will offer an immediate solution in the form of micro-clouds.
The data and insights in this article are sourced from NewtonX experts. For the purposes of this blog, we keep our experts anonymous and ensure that no confidential data or information has been disclosed. Experts are a mix of industry consultants and previous employees of the company(s) referenced.