Bryan Cantrill: How Kubernetes Broke The Aws Cloud Monopoly

Did Kubernetes break up Amazon Web Services unreality party?

As nan co-founder and CTO of unreality autochthonal hardware institution Oxide Computing Company, Bryan Cantrill isn’t 1 to awkward distant from speaking his mind. A caller wide-ranging question and reply connected nan  Pragmatic Engineer podcast, hosted by Gergely Orosz,  he offered a position connected really nan merchandise of Kubernetes was basal for fueling nan maturation of nan Google Cloud Platform and different nascent unreality providers, erstwhile AWS was relentlessly dominating nan market.

“I deliberation that portion of that first attraction to Kubernetes is that group wanted to get immoderate optionality astir their unreality and they felt locked into AWS,” he said.

“There was a play of clip erstwhile I felt for illustration successful bid to beryllium successful nan cloud, you had to instrumentality each AWS API. It was thought that Google and Azure could ne'er compete pinch AWS because they could ne'er beryllium API compatible.”

Cantrill has had a knack astatine being astatine nan correct spot astatine nan correct time. After graduating from Brown University successful 1996, he took a occupation pinch Sun Microsystems, which was past nan spot that companies needed to bargain servers from to get connected nan Internet. There, he helped create DTrace, a powerful observability tool, primitively for Sun’s Solaris operating system, a Unix variant.

When Oracle bought Sun Microsystems, completing nan acquisition successful 2010, Cantrill, for illustration galore Sun techies, fled soon after. He was 1 of nan co-founders of Joyent, a unreality work that was marketed arsenic a engineer-savvy replacement work to nan bigger clouds. Cantrill went connected to recovered Oxide, which sells servers to tally soul unreality services.

The AWS Juggernaut

From 2010 to 2014, led by Andy Jassy, AWS was going done a play of “relentless execution,” Cantrill recalled.

In 2010 nan institution was still seen arsenic a “visionary” by Gartner. But by 2014 — nan first twelvemonth Amazon collapsed retired AWS income successful its net study —  nan institution brought successful $4.6 billion successful revenue, and had astir 5 times nan capacity arsenic each nan different unreality providers that Gartner tracked.

It seemed unbeatable. Each twelvemonth astatine the AWS Re:Invent personification conference, they institution announced value cuts, and a staggering number of caller services. Vendors retired connected nan show level would abruptly find themselves competing pinch immoderate caller work AWS conscionable announced (and, fable has it, moreover immoderate that packed their booths and near mid-conference).

“If you were a competitor to AWS you were for illustration dreading Re:Invent because present comes different value cut. If you are a partner of AWS, you’re dreading Re:invent because present comes nan announcement of a caller work that competes pinch what you’re making,” Cantrill recalled.

Financial reporters will retrieve that, successful its quarterly filings earlier 2014, Amazon did not break retired its net specifically for AWS. Typically, companies obfuscate these numbers to hide anemic performance. AWS was really doing nan opposite,  Cantril believes. It was hiding its unreality gross because it was truthful good, and reinvesting backmost into nan unit broadside of nan business.

Joyent was competing pinch AWS successful offering a nationalist unreality astatine that time, and truthful nan institution saw really lucrative nan business could be.

“We ran a nationalist unreality and we knew nan economics of a nationalist cloud. The margins were good.”

In fact, respective of Joyent’s astir salient customers were retailers successful nonstop title pinch Amazon, such arsenic Walmart, and presumably not willing successful utilizing a competitor’s unreality service.

“Retailers are like, “If you deliberation I’m going to return my dollars and walk them connected AWS truthful Amazon tin spell to warfare pinch me, [then] nary convey you,” Cantrill surmised.

The Power of Running Clouds In-House

The thought of a company, specified arsenic a Fortune 500 organization, building its ain “private” unreality in-house seemed for illustration an charismatic replacement to many, particularly arsenic nan customers saw their AWS bills grow.

Joyent had unfastened root package for moving a cloud, namely nan SmartOS, a version of OpenSolaris, that allowed businesses to cobble their ain unreality strategy utilizing commodity hardware. Samsung purchased nan institution successful 2016, purely to powerfulness its ain unreality operations, Cantrill said.

It whitethorn person been a savvy move for Samsung, though nan acquisition near different businesses pinch 1 less action for taking complaint of their ain clouds.

Other options astatine nan clip included Mesosphere and Docker Swarm. Many thought nan champion measurement to compete pinch AWS was to replicate AWS’s ain APIs. This was done successful an unfastened root task called Eucalyptus, which was purchased by HP successful 2014.

Enter Kubernetes

In this chat pinch Orosz, Cantrill makes nan statement that it was really nan introduction of Kubernetes instrumentality orchestration motor successful 2014 that democratized nan unreality market, moreover if it wasn’t that wide utilized astatine first (It’s now nan 2nd largest opens ource task aft Linux).

At nan time, Google offered its unreality services, but internally, they were tied to nan circumstantial bits of hardware. Kubernetes came astir arsenic a task to instrumentality “cluster-based scalable web services,” according to one of nan original papers describing nan technology. The emergence of Docker astatine that time allowed of giving each workload its ain IP address, acknowledgment to containers.

The committedness of Kubernetes is that it provides nan expertise to proviso basal infrastructure, but not beryllium tied to a circumstantial hardware level itself. No longer would you request to constitute to nan AWS APIs. Applications could beryllium built to nan Kubernetes APIs and, successful theory, beryllium capable to tally connected immoderate unreality provider, achieving a shape of operational unreality neutrality.

“Multi-cloud didn’t really exist, I’d argue, earlier Kubernetes,” Cantrill said. Indeed, galore of nan early contributors to nan task were motivated by this perfect of unreality neutrality.

Why Did Kubernetes Open Source Kubernetes?

Why Google unfastened originated Kubernetes, which was an offshoot of its soul Borg platform, successful nan first spot remains a spot of a mystery.

Of course, there are benefits that travel pinch unfastened sourcing immoderate project, specified arsenic much contributions to nan codification guidelines and security.

But releasing an unfastened root task to promote unreality neutrality besides made competitory sense, astatine slightest from Google’s P.O.V.

“I deliberation that nan statement that group were making internally was astir encouraging unreality neutrality because ‘We are nan ones that person thing to win,'” Cantrill said.

Google moreover gave nan Linux Foundation seed money to motorboat nan Cloud Native Computing Foundation in 2015 to independently manage Kubernetes and related unfastened root technologies.

Nonetheless, nan Google Cloud Platform has since recovered its footing successful nan marketplace. GCP unsocial now brings successful astir $60 billion successful gross to Alphabet annually.

AWS is still nan apical provider, pinch astir a 30% stock of nan unreality marketplace, though nan marketplace itself has ballooned successful size to beryllium a trillion-dollar marketplace today, pinch a wider assortment of participants that person gained marketplace share.

“Is Kubernetes to convey solely? No, but I deliberation it’s played an important domiciled for sure,” Cantrill said.

Enjoy nan full talk here:

Group Created pinch Sketch.