Case Study: Kiwi.com achieves 900K reads/sec and 75% lower running costs with ScyllaDB

Kiwi.com Takes Flight with Scylla

Kiwi.com is an online travel platform that builds complex itineraries (Nomad) by mixing flights from many carriers; it handles heavy demand (about 90,000 daily queries and 25,000 seats booked per day) against a combinatorial dataset of roughly 7 billion flight combinations that refreshes rapidly. After struggling to scale on Postgres and then Cassandra—while aiming to run globally on bare metal and migrate to GCP—Kiwi.com evaluated alternatives and selected ScyllaDB as a drop-in replacement for Cassandra.

Kiwi.com and ScyllaDB ran a proof-of-concept and production migration to GCP and OVH multi-region clusters; ScyllaDB delivered dramatically better performance (21 nodes at 900K reads/sec vs Cassandra’s 100 nodes for 40K reads/sec with RF=4) and cut running costs to roughly 25% of Cassandra. The ScyllaDB deployment now supports the 7 billion-entry dataset (11 TB in replicas) with about 700K writes/sec and 500K reads/sec, simplified operations by removing custom data-layer code, and enabled better prioritization of production traffic versus analytics.

Kiwi.com

Martin Strycek

Engineering Manager