Is Software Eating Networking? Facebook Says ‘Yes’


Thanks to Linux, open source has become the de facto development model for a majority of enterprise software projects — and that’s quickly becoming true in the networking space, as well.  

Networking is playing a pivotal role at Facebook, for example, which has written a lot of open source-based software to automate, roll out, and monitor its massive networking infrastructure, said Omar Baldonado, director of engineering at Facebook, in a keynote presentation at Open Networking Summit (ONS) last month.

But one question keeps popping up: how to make money from open source. Peter Levine, general partner at Andreessen Horowitz, a venture capital firm, addressed this question during his ONS keynote. Open source must go beyond Red Hat’s support business model and come up with new business models, he said.

We’ve collected some highlights from these two keynotes, below, and videos are available of more than 20 keynote and plenary sessions from ONS 2016 — watch them now.

Open Source Needs a New Business Model

Although open source has been a tremendous technological innovation, the future of open source is going to require business innovation. Levine said that open source as a service is a great business model and cited the example of companies like GitHub and Databricks that have built businesses around open source. They are not directly monetizing from open source the way Red Hat is.

Another business model, according to Levine is “open core” in which companies build proprietary software on top of open source and sell it as a subscription or traditional licensing model.

Levine then said that the biggest open source company in the world is not what everyone thinks; it’s not Red Hat. Facebook releases a lot of its work in open source and has come up with a business model around advertising — a completely different monetization mechanism.

His point was that companies need to think outside the box to make money from open source.

Developers. Developers. Developers.

Levine also made the interesting point that in today’s world developers are becoming more and more powerful. In the early days, you could buy a piece of hardware and give it to developers to write software for it. Now applications are becoming the lead instead of the underlying hardware or infrastructure. 

“If I am a systems company or a networking company and I am going to sell my product, then developers are a very interesting place to start in terms of where products are getting adopted and and how they are getting used,” said Levine. And, as software becomes more and more important, almost every company becomes a software company.

Levine said that every time a disruption occurs, there is an opportunity for new folks to come in and the incumbent base needs to figure out how to actually get there. “Some of this transformation makes it nontrivial to jump over the fence to the other side,” he said.

He gave the example of how virtualization disrupted the server market and it was not the incumbents like Dell, HP, and IBM that benefited; it was software companies like VMware.

We have already seen this change occurring in data centers, and the same is going to happen in the networking space, so the challenge for traditional networking companies is how to make this transformation — which will also cannibalize their own base.

Back to the Future

As an investor, Levine says he doesn’t care about where the hockey puck is; he cares about where it’s going. The gist of what he said was: Cloud computing is not going to be there forever. It will change, it will morph, and he believes that the industry will go from centralized to distributed and then back to centralized. He said that the cloud computing model is going to disaggregate in the not too distant future back to a world of distributed computing.

This is already happening. Whether it’s Tesla or VR handsets, processing is happening locally and not in the cloud. Levine believes that peer-to-peer operations at the end point will become the trend, although it won’t be the same peer-to-peer that we have seen in the past. These peers will be your cars, planes, and boats. It was a very interesting and thought-provoking perspective.

Both Hardware and Software Matter: Facebook

Levine’s keynote was followed by a presentation from Omar Baldonado, director of engineering at Facebook, who said that both hardware and software are equally important to networking.

He gave an overview of how networking and software are playing a pivotal role at Facebook. No matter what Facebook component you talk about — whether it’s outside-facing products like WhatsApp, Messenger, Groups, Instagram, or Oculus, or internal projects like big data analysis, live streaming, and machine learning — they all run across the same infrastructure and network. The networking teams get involved in everything. Baldonado said they have written a lot of software to automate, roll out, and monitor this massive networking infrastructure.

Facebook’s networking team upgrades the software every week, and Baldonado wants this practice to be adopted elsewhere.This way operations teams don’t have to wait for a whole year and then worry about the 10,000 new features and changes that they have to deal with in upgrades.

The traditional approach to  “upgrading” becomes a massive event in itself, which also increases the risk of failure. One or more of those 10,000 features is going to break something, and then it will be hard to find the culprit.

Baldonado also emphasized the policy of “fail-fast” rather than “fail-proof.” He said it’s a tall order to create a 100 percent fail-safe network. He said that we should focus on finding the problem quickly and then automatically resolving it through software. If you assume that different parts of your network are going to fail, your design should accommodate all those failures. You can’t find a perfect switch that never fails. Something will fail, so you should build a system that can detect it and fix it as soon as possible.

He also talked about NetNORAD, which detects network interruptions automatically and mitigates them within seconds — all through software. And, he noted that Facebook has contributed code to some of that work through its open source participation.

But, for all of the amazing things that they are doing with software, Baldonado said, hardware is still just as important. And, that’s where Facebook’s Open Compute Project (OCP) comes into the picture.

“I heard that software eats networking, but software needs to run on something,” he said. And, he noted that OCP shows there is a very rich base of hardware innovation that is waiting for more software to be written to control, manage, monitor, and orchestrate it.

Commenting on the future, Baldonado said that, at Facebook, they want to continue working with the new networking ecosystem, sharing what they do and learning from what other people are doing, too. “We keep on coding!”