How Criteo scaled 290M QPS and cut its server footprint by 78%

Criteo, a company serving over 700 million users daily with personalized ads, has scaled its real-time engine to handle 290 million key-value queries per second (QPS) without increasing costs. The company replaced its complex stack of Couchbase and Memcached with Aerospike's patented Hybrid Memory Architecture (HMA), which decouples index storage from data storage. This change allowed Criteo to consolidate two systems into one, simplify operations, cut its server footprint by 78%, maintain sub-millisecond latency at global scale, and reduce operational overhead. The company achieved these improvements while adopting a Kubernetes-native deployment with the Aerospike Kubernetes Operator, which enabled automatic node recovery and rebalancing via Kubernetes. By making multi-bin optimization and avoiding over-engineering, Criteo was able to reduce index memory usage, improve access efficiency, and maintain consistency and throughput under stress tests. The company's experience demonstrates what becomes possible with the right storage architecture for real-time systems with massive QPS and tight SLAs.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.