It’s a no brainer that eBay and other big online retailers need lots of computing power to run their sales efforts. But the niceties of building a tech foundation that is both powerful enough for their immediate needs and flexible enough to meet changing demands tomorrow, is tricky.
It’s also worth noting that given the competitive nature of the retail world, companies like eBay (EBAY) on the e-commerce side and Walmart (WMT) with feet in both brick-and-mortar and e-commerce camps, are not turning to Amazon Web Services for their infrastructure needs, given that parent company Amazon.com (AMZN) is seen as the ultimate enemy.
The need for powerful-but-adaptable computing is why eBay, which claims more than 800 million sales listings and which has always relied on its own internal tech expertise, continues to construct its own massive computing capabilities using OpenStack, a set of freely available modules for computing, storage and networking. It also uses Docker container technology, which is a modern way to package up all the components of a given application so that it can run almost anywhere, whether in a company’s own server room, or a private company-owned cloud, or even a public cloud owned by a third party a la Amazon (AMZN) Web Services, Microsoft (MSFT) Azure, or Google (GOOG) Cloud Platform.
Ebay also relies on Kubernetes, a technology that developers can use to ease deployment and management of lots and lots of those Docker containers, a task that gets complicated in a hurry. Kubernetes was pioneered by Google (GOOG) but is now backed by a bevy of tech giants including IBM (IBM) and Red Hat (RHT).
By making use of this modern array of tools, eBay can make sure its computing resources can adapt, said Debashis Saha, eBay’s vice president of commerce platform infrastructure.
“We have a humongous amount of infrastructure that we need to provide in a modern way, with scalable microservices and machine learning,” Saha told Fortune.
Let’s break that down. Microservices are lots of little independent software applications that talk to each other via a standard connective conduit called an application programming interface or API.
This is very different from the gigantic software applications that big companies used to use. They took forever to roll out, were hard to maintain, and really hard to customize or change mid-stream. Microservices are roughly analogous to a flotilla of fast speedboats working in tandem as opposed to an older application that’s comparable to a giant ocean liner. Both may do the job, but the speedboats are faster and more agile.
Machine learning, or artificial intelligence, is technology that enables computer software to learn on its own, by plowing through data and refining its responses to a given problem based on that data. Self-teaching AI systems thus can adapt based on the data they parse.
Get Data Sheet, Fortune’s daily newsletter about the business of technology.
OpenStack, Docker, Kubernetes are all tools Ebay and other companies can use to, in Saha’s words, make overall infrastructure programmable: It can be made to adapt to changing needs without always requiring a big hardware refresh and downtime.
Because Ebay has been around the block—it was founded in 1995—it had built its own infrastructure to suit its needs over time, but that infrastructure was rigid with home-grown applications were “stitched together in a proprietary way,” Saha said. That meant changing one application could have negative ripple effects on others, which can take time to track down and fix.
By using Kubernetes and Docker, Ebay was able to to create clusters of computing power in its data centers that can grow or contract with workloads. “Marrying Kubernetes with OpenStack was very important because Kubernetes needs a set of servers to manage. Putting Kubernetes atop the OpenStack foundations means that clusters can be moved if needed.” The goal here is what Saha calls programmable infrastructure.
If all you need is simple compute, storage and networking you can aggregate them without need for Kubernetes. But if you need to run flexible applications that can be dynamically run and scaled Kubernetes is key, Saha said. It represents the “best practices of the application layer.”
This is a work in progress that started with eBay’s OpenStack projects three or four years ago, and Saha expects that Kubernetes capabilities will be available up to all of the company’s developers by the middle of this year.
Ebay and Paypal part company:
So this is all great for eBays’ developers who get to use the latest and greatest tools and open source software—expertise which makes them more valuable. But for the ultimate user, in theory this means a shopper will be able to find what he’s looking for faster and more easily and get a more personalized experience. Use of machine learning means that, based on the shopper’s past purchases or what he’s browsed, the system itself can help target searches more accurately, tailored to the unique user.
Faster, better searches, and a better overall experience can mean money in the bank for eBay, and an easier time for Marie Kondo readers auctioning off the contents of their closets.