It’s 5:15 p.m. and 60 degrees on a rainy November 30 at your home in Miami. You go to Allrecipes seeking inspiration.
Behind the scenes, Cortana machine-learning technology in Microsoft’s Azure cloud service figures, “Hmm. It’s just after Thanksgiving, in the late afternoon, and chilly for Florida. Bet there’s a lot of turkey leftovers, and it’s nearly dinner time. I’ll suggest the ‘Thanksgiving in a Dish’ casserole.”
That’s how Allrecipes sees its new technology unfolding within the next several months — just part of the 19-year-old company’s major migration to Azure, said CTO John Keane in an interview today.
“We want to be the Pandora for food, to offer customized suggestions based on preferences you’ve expressed, recipes you’ve made in the past, and where you live, the time of day, the season and even the current weather,” he said.
Seattle-based Allrecipes, which says it serves up 95 recipes per second in 24 countries and 13 languages, experiences wildly varying web traffic. It gets hammered with a month’s worth of visits on the day before Thanksgiving and is extremely busy during December and on Sundays, but on other days and at many times of the year, many of its 500 servers — some in Seattle, some in Boston — sit idle, Keane said.
Wasted resources like that impelled the company to move its entire production infrastructure to the cloud. Using mostly Platform as a Service (PaaS) from Azure, “we can expand and contract as necessary, sizing up automatically as traffic increases,” he said. “That’s really a great capability for us.” The move is scheduled to occur within the next two weeks.
At least for now, some computing is being kept in the data centers, including SQL Server databases and legacy apps that run analytics to provide sales insights. Moving SQL Server to the cloud isn’t essential, as the size of the database doesn’t vary much, Keane said. The age of the legacy apps makes them harder to move, he said.
With Amazon Web Services already in limited use at Allrecipes, why did the company choose Azure?
“It was a great fit for us,” Keane said. “Microsoft’s development tools are a natural fit for the C# we use, and it was easy to integrate our build environment and push to the cloud. AWS has a great set of tools, but it feels more fragmented, and you need a certain amount of skills. It wouldn’t have been as easy a move.”
Beyond that, he said, was the personal element. “The Microsoft reps honestly were much more willing to help us get up and get going. Amazon is a good company, but the Microsoft team was present, engaged, forthright and wanted to help. The willingness to form a partnership was important.”
Moving to the cloud could save 20-30 percent, he estimated, stressing those figures are tentative.
Though greater efficiency is unquestionably a big benefit, especially with a DevOps team of only about 20 people, it’s the machine-learning personalization that seems to excite Keane the most.
“One of my passions has been to create a highly personalized food experience that shows you things you might not have thought of,” he said. “It will be a fun and very compelling experience.”