Built to host WordPress, now carrying three billion phones: Mark Collier on where OpenStack went
Full disclosure, which I also gave on camera: Mark Collier and I have known each other for about a decade. I represented Yahoo when the OpenStack Foundation was set up in 2012 and sat on its board for three terms. Mark left Rackspace to become the foundation's COO and has been there ever since. So this was two people who were in the room comparing notes on what happened to the thing, ten years after the first release.
What happened is not what anyone planned. OpenStack started in 2010 as a joint project between Rackspace, which at the time was one of the two big cloud providers along with Amazon, and NASA, which had a lot of data and wanted automated infrastructure software built in the open. The idea was to take compute, storage and networking in a data center and automate them, which is what we now call cloud computing. Two years later the foundation gave it a neutral home. Over the decade more than eight thousand developers in over a hundred countries have contributed. Mark's explanation for why it worked is the one I have come to believe myself: the smartest engineers are never in one country or one company.
Nobody at that table was thinking about telephone networks.
When we started it, we had no idea it would ever be used for mobile networks. None. And that was completely unfathomable to us. We were thinking about servers and a data center for web applications, as typical things people want to put in a web hosting environment, WordPress, whatever you want to host your website on. But it turns out that processing, storing, moving data in an automated way is exactly the kind of use case that networks have. Almost every single mobile network in the world is now running OpenStack. I think nine of the top ten are now running OpenStack, which represents like three billion subscribers.
I wanted to know where in the network it runs, because that has changed too. It began in the big centralized data centers it was designed for. Then 4G and now 5G pushed the latency requirements down, and the heavy lifting moved outward. AT&T, Mark said, runs something like two hundred OpenStack clouds inside its network and wants thousands, which turns the word cloud upside down: not one enormous facility but many small ones, as close to the subscriber as the carrier can put them. His reason was the one physicists give. The speed of light is the speed of light, and until the quantum breakthroughs arrive you move the compute to the user rather than the other way around.
The newest example is the radio access network itself. Verizon announced it is using StarlingX, a project the foundation shepherds that combines OpenStack with Kubernetes, for its 5G RAN. Functions that used to live in fixed proprietary boxes at the tower are becoming software, and the carriers are adopting open source for the usual reasons: lower cost, faster updates, and no rip-and-replace when the architecture shifts. Mark was careful about the word edge. In this world it usually means the point where the towers meet the carrier network, though there are towers running OpenStack now too. It does not mean the handset.
Which is why the foundation is renaming itself. Starting in 2021 it becomes the Open Infrastructure Foundation, still working on OpenStack but also on StarlingX, Airship, Zuul and Kata Containers, with the next decade of infrastructure as the brief. A lot of that brief comes back to the carriers, and to the question Mark left open at the end: as the bottlenecks move around, what runs all the way out at the tower?
I came away with a thought I did not have in 2012. You do not get to decide what your software is for. You decide whether it is open enough for someone you have never met to find the use you missed. The people who found this one run three billion phones.
From Lincoln Shorts, the interview series I hosted for Lincoln Network, recorded 17 December 2020.
