From DevOps to No-Ops: El Reg chats serverless computing with NYT's CTO
Or how to stop worrying about infrastructure and love lock-in
Posted in DevOps, 30th November 2017 08:02 GMT
DevOps, a combination of development and operations, may have to be rethought because ops is on the outs.
In recent years, those who develop applications and those who manage the machines hosting the apps have tended to tended to be distinct. Even recently, as physical machines have given way to virtual machines and containers, there's still a separation of concerns. At least in large organizations, the person creating application code probably isn't the one overseeing Kubernetes clusters, network infrastructure, or dropping by the data center in the wee hours.
Serverless IT architecture doesn't eliminate those concerns but it does shift a lot of them to Amazon, Microsoft, Google, or other cloud service providers.
"There's the no-ops dream," said Nick Rockwell, chief technology officer at The New York Times, in a phone interview with The Register. "Many people snicker, but I think for many applications, we can get very close."
Rockwell is bullish on serverless, yet another term that should be revisited, but for being a misnomer rather than in response to industry change. Serverless IT requires servers; it's just that those using them don't have to deal with them because the service provider does so.
"I've come to feel this could be bigger than cloud," said Rockwell, who has written on the topic and said as much at conferences. "You don't have to think about scaling or availability."
At The New York Times, said Rockwell, Google Kubernetes Engine (GKE) is the main platform for apps. And the newspaper's engineers have been using Google services such as Cloud Pub/Sub, Cloud Spanner, Cloud Datastore, App Engine, Container Engine, and BigQuery.
However, Rockwell isn't all that keen on containers; he'd rather rely on Google's App Engine, once referred to as a platform-as-a-service, and now fitted for current fashion with the serverless sobriquet. His goal, he said, is for all new apps to be serverless by 2019.
"Serverless" has tended to be used to refer to cloud services that execute functions on-demand like Lambda, Azure Functions, and Cloud Functions rather than more complex applications. But between Google App Engine, Amazon's newly announced Fargate – containers without the management – and its Aurora Serverless database, restricting the concept to functions no longer seems adequate.
As Rockwell defines the term, serverless means any platform that abstracts scaling and reliability. But in all likelihood, serverless will become as overused and meaningless as AI.
Whatever you call the path toward ignoring infrastructure – serverless or folly – there's less ops in the mix. Dev(ops), anyone?
"We're in the process of a transition where you're seeing the ops team get smaller," said Rockwell. "We went from hosting our own Hadoop cluster to BigQuery and Pub/Sub. We went from a giant admin headache for Hadoop to no-ops."
Rockwell argues that the dev side of things matters more because it demands more resources. "As a CTO, I spend so much more money on development than infrastructure," he said. "If serverless helps optimize development, that has a bigger impact on the organization."
We survived today's Amazon news avalanche to bring you this: Yes, a managed Kubernetes service will be a thingREAD MORE
Not everyone is sold on the idea. By Rockwell's own account, his presentation on the subject at O'Reilly Media’s Velocity Conference last month was not well received.
But others in the industry see serverless – managed? – IT making gains. Via Twitter in August, Subbu Allamaraju, Expedia's veep of cloud, said: "Serverless patterns are pulling the rug from underneath container cluster managers faster than the latter becoming industrial grade."
When The Register spoke with Alex Polvi, CEO of enterprise container management biz CoreOS, earlier this month, he too suggested serverless – aka PaaS – has a bright future, once the open-source community delivers an alternative to the proprietary lock-in of Amazon's Lambda.
Rockwell didn't see that as much of an issue. He said lock-in is a concern but it's a trade he's willing to make given the benefits. "I will gladly lock myself into the best platform because the benefits are so significant," he said, suggesting that the innovation in IT infrastructure right now is being driven by the big cloud service providers and standing on the sidelines means missing out.
"If you don't accept lock-in, you won't get the most out of these platforms," he said.
Who'd want to see a pair of handcuffs go unused? ®