When you think of retail giant Wayfair, known for selling rugs, furniture and décor at bargain prices, the cloud is probably not the first thing that comes to mind. But perhaps it should be.
Matt Ferrari, Wayfair’s head of martech, data, and machine learning in its Engineering and Product division, told Silverlinings it recently completed a “lift and shift” of all 6,000 of its enterprise applications to the cloud and is halfway through the process of refactoring them to be cloud-native. He dished on how it went about the process and what he would do differently if given the chance to start again.
According to Ferrari, Wayfair’s cloud journey started about four years ago, when it began moving back office applications – think email, G-Suite and other non-customer-facing applications – to the cloud. Toward the end of 2019, it kicked its migration effort into high gear, and by Q3 2022, had moved all of its databases and applications to the cloud. About 95% of its apps are in Google Cloud, though a small portion is in Microsoft Azure.
The why behind it all? Focusing company resources where they could make the most difference for customers and suppliers, he said.
“Wayfair has 3,400 engineers across the ecosystem. We didn’t want them thinking about servers or storage, we wanted them thinking about the customer experience or how to really delight a supplier,” he explained.
“We wanted to change the way that engineering thought. Wayfair was not going to differentiate in the field of compute, but Wayfair is able to differentiate in the field of machine learning, generative AI, those kinds of things.”
The moving process
There are a couple different elements to Wayfair’s cloud transformation. Ferrari said the vast majority of its apps are traditional and backed by a relational database like Microsoft SQL. Now that they’re all in the cloud, though, Wayfair is working to decouple its apps from SQL and moving to alternatives like Alloy DB and Postgres. The idea behind this is to enable the company to scale more easily. Ferrari noted the ability to scale on demand is particularly important for a web-based retailer like Wayfair that has to respond to seasonal consumer buying patterns.
He said it is also working to move from a monolithic architecture to a microservices architecture. There, the concept is “rather than having a spaghetti string of code, let’s have code that has a single purpose and so we know what when we make a change to a feature or function we know exactly where to make that change.”
And finally, Ferrari said it is moving away from virtual machines to a Kubernetes-based serverless architecture. He explained that while it is in the process of refactoring its apps, many still run on virtual machines. But he noted virtual machines run using operating systems. So, by eliminating that operating system from the equation, the company is removing a potential point of failure in its infrastructure. The move to serverless also comes with security benefits, reducing the “blast radius” of any threats, and, again, offers greater ability to scale as needed.
The 'power of the public cloud'
Ferrari started at Wayfair in 2020, after the company had already made the decision to move to the cloud and began its journey.
Asked what he would do differently if given the chance to start over, Ferrari said he would have focused less on pursuing a “lift and shift” transition and more on application decoupling and modernization upfront.
“Lift and shift makes sense in the short term, but in order to really leverage the power of the public cloud, moving to cloud-native technology sooner is certainly better,” he said.
In terms of other tips, Ferrari said it’s absolutely critical for enterprises to properly tag workloads so they can be associated with the right internal billing codes.
“When you don’t have that, all of the sudden you get a ginormous or scary bill from a public cloud organization and you don’t know how to parse it. A public cloud invoice is as confusing to a novice as the first time they log into a public cloud portal,” he said, adding bills typically include a log of every cloud service used. “If you don’t properly tag your assets, you very clearly can’t let’s say chargeback or do cost and show-back to individual organizations in your enterprise.”
His final advice for success? Build and embrace a culture of change.
“Organizations that move to public cloud and they just treat it as a data center or just another network or a storage device are missing the point,” he concluded.