Criteo relies on time-critical applications to serve up advertisers’ content to potential consumers. They perform matches 950 billion times a day with 50 ms response times. With growing demand among advertisers to reach more consumers and grow online sales, data loads and requests accelerated the demands on Criteo’s data infrastructure.
The system was a large global deployment spanning on-premises data centers in Asia, Europe, and North America. It needed to handle real-time decisions at 250M transactions per second at peak, involving data all over the world at sub-millisecond latencies. Use cases included managing images, user data, campaigns, and AI models.
Criteo’s original solution used Memcached and Couchbase. Criteo was stuck on an outdated version of Couchbase to avoid scaling and budgetary limitations that came with newer or enterprise versions. They soon began to struggle to keep up with the data demands.