Case Study: MediaMath achieves massive scale, sub-10ms latency, and 99.9999% uptime with ScyllaDB

MediaMath Innovates AdTech at Scale with Scylla

MediaMath, a leading independent programmatic ad tech company, faced a scaling challenge: massive data volumes from media touchpoints, very high transactions per second, and real-time bid-matching and segmentation requirements. Their initial Cassandra deployment — up to 60 nodes on AWS — created heavy operational overhead (two to three full‑time SREs for maintenance, manual restores, compactions and JVM tuning). Seeking a compatible, higher-performance alternative, MediaMath evaluated and adopted ScyllaDB (Scylla) as a drop‑in replacement for Cassandra.

ScyllaDB’s C++ implementation and Cassandra compatibility enabled a smooth migration with no tooling rework, letting MediaMath reduce cluster size and operational effort. Today MediaMath runs 17 Scylla nodes on i3.metal instances, handling ~200,000 events/sec and ~1,000,000 reads/sec with read latency consistently under 10 ms and 99.9999% availability during peak periods. The move lowered operational overhead, freed engineering capacity, and allowed MediaMath to increase segmentation data retention and deliver new client value without raising prices.

MediaMath

Knight Fu

Director of Engineering