Case Study: Grab achieves 40K ops/sec low-latency, cost-efficient stream aggregation with ScyllaDB

Grab Hails Scylla for Performance and Ease of Use

Grab, a leading Southeast Asian superapp used for rides, food and payments, faced a critical low-latency data challenge: aggregating multiple Apache Kafka streams in near real time for millions of daily rides. After Redis showed CPU spikes and operational limits, Grab evaluated alternatives and selected ScyllaDB’s Scylla NoSQL database as their aggregation metadata store to meet heavy write and read demands.

ScyllaDB was deployed in production on a compact 5‑node AWS i3.4xlarge cluster and delivered Redis‑level performance while handling peak loads of about 40,000 operations per second. The ScyllaDB solution cut projected infrastructure costs by more than threefold versus other options, eliminated hot‑partition issues, improved read/write latency, simplified cluster operations compared with Cassandra, and provided fast vendor support—prompting Grab to expand ScyllaDB to additional use cases.

Grab

Aravind Srinivasan

Software Engineer