Intel's Nick McKeown: 5G means much more than smartphones

The next killer application of 5G will open the floodgates of innovation for enterprises.

Nick McKeown, Senior Vice President and General Manager of the Network and Edge Group at Intel Corporation

Your smartphone can go from egg timer to real-time language translator to augmented reality game console thanks to its programming capabilities, which can turn your wildest ideas into “an app” to help you achieve your goals.

Modern computing infrastructure, including cloud data centers, internet and cellular networks, and edge computing, makes your phone capable of so much. But there’s still one part of this infrastructure that’s not fully programmable: the network itself, which means users can’t turn their wildest ideas into reality.

So what's the reason? When the Internet industry began to develop in the 1990s and early 2000s, it was caught in a dilemma of too many standards and regulatory stakeholders and the network industry being too complacent. The Internet industry, which should have been open, simple, fast-growing and flexible, fell into a state of rigidity - developing at an extremely slow pace. Of course, although the speed of the Internet has become faster, it is difficult for the Internet to provide users with more reliable, secure and practical services due to the limitations of standards and chips.

For fifteen years, I have been eager to solve this problem and improve the Internet industry, make it faster, and enable faster development. When I refer to the Internet, I mean the broad network: the networks in our homes, cellular networks, Wi-Fi networks, enterprise networks, the public Internet, and the networks inside cloud data centers.

This desire drove me to be a professor at Stanford and to start several successful networking companies before I joined Intel last year. My goal has always been to challenge the networking industry to think more about software-driven infrastructure.

In the past, all the capabilities of a network were constrained by standards and equipment manufacturers, who had little incentive to change. They believed this was the only way for the network to achieve expected performance, control costs, and ensure power efficiency. But now, things have changed.

A real-life example is happening today in Japan. A recent study of 5G networks found that the provider with the fastest download speeds — more than 40 percent faster than other providers — used a network built by Intel customer Rakuten. Rakuten’s virtualized network runs on Intel Xeon processors and uses our FlexRAN software.

It is worth noting that Rakuten Mobile is not a telecommunications company, but an e-commerce and Internet service company with 1.5 billion global members. However, Rakuten is able to build a 5G network using software based on existing infrastructure (mainly used to provide dozens of online services).

Many companies with warehouse-style data centers—such as Google, Amazon, Facebook, and Microsoft—are also moving to programmable networks. The main reason for these companies' transformation is the need for faster speeds, but the transformation also gives them greater flexibility in programming.

Let’s take a closer look at the need for speed in today’s businesses. As a simple example, if we draw a vertical line across the United States and look at all public Internet traffic from left to right and then back to left, the size of the Internet capacity in the United States (also known as the Internet’s binary bandwidth) is less than the traffic between a few hundred servers in a modern data center, which typically contains tens or hundreds of thousands of servers and is very large (hence these companies are often called “hyperscale companies”).

If one of these companies wants a data center that's faster, more reliable, and more secure than its competitors, it can't just buy the same old fixed-function network boxes. To introduce innovative ideas that differentiate it from its competitors, each company needs to program these devices itself.

The natural next step for these companies is to take chips that are customized to their needs, programmable, and have differentiated features. I’ve been involved in developing network chips with this goal in mind. Over the past decade, I’ve watched as companies want more control over how individual packets are processed, and they do interesting, innovative, and sometimes crazy things that I never thought of—and their competitors never thought of. Today, networks in different data centers work differently because they add their own differentiated features to gain a leg up on the competition.

If you look at it a little bit, the whole system—the computers, the storage, the network—is becoming one big distributed system that you can program to do whatever you want.

Our job at Intel is to provide our customers, especially their software developers, with the world’s best programmable platforms. As infrastructure shifts to software in the cloud through the internet and 5G networks, all the way to the intelligent edge, our job is to make it as easy as possible for these developers to develop with our hardware to realize their new ideas.

By liberating hardware functions to software, our customers and developers can improve the functions themselves faster than ever before. Because once the functions are fixed by the hardware carrier, innovation will not only be slow, but also inevitably limited by the imagination of the hardware manufacturer.

However, if you move it into software, you open it up to a much wider audience, meaning a large number of developers can come in and try out their ideas. More importantly, you've handed the keys from the people who make the hardware to the people who own and operate large network systems for a living. Only they know how to operate at that scale; only they can write the software that determines how their systems should work.

As the global computing fabric finally becomes programmable, it will change everything—and open the way for untold innovations.

For example, my colleague Raja Koduri recently argued that the Metaverse could be the next major platform in computing after the World Wide Web and mobile Internet. To realize this vision, we need orders of magnitude more computational and communication capabilities, accessible with lower latency across multiple device form factors. All of this is more easily achieved through a more composable and programmable infrastructure.

Fully programmable infrastructure will also enable a wider distribution of intelligence. For example, it enables data processing capabilities to be closer to where the data is generated or consumed, which is what we call the edge.

At the edge, our customers are already deploying a lot of AI inference in their premises, analyzing video from cameras to monitor inventory, measure customer traffic, and identify anomalies in production. The applications for inference are already widespread and will grow rapidly, and we will see a huge transformation in factories, retail stores, and hospitals. As AI inference grows, developers will need open, programmable models so that they can achieve whatever they want with their innovative new applications, rather than being limited to a single solution. For this reason, we have seen rapid growth in our successful OpenVINO inference platform. When combined with 5G, we believe this is the next killer application.

I can't wait to see what new ideas come out next, especially the crazy ones.