Case Study: Allegro achieves 4x throughput, 7x lower response times and major cost savings with ScyllaDB

Allegro Puts Scylla to the Test

Allegro, Poland’s largest e‑commerce marketplace with more than 70 million items, 14 million monthly users and about 750,000 purchases per day, was running Apache Cassandra across 20 clusters (250+ nodes) but faced frequent garbage‑collector pauses, slow maintenance tasks (restarts, repairs, joins, cleanups), and recurring latency spikes despite extensive Java tuning. To address performance and cost concerns, Allegro evaluated ScyllaDB (Scylla) as an alternative.

In production‑traffic tests using mirrored clusters (two datacenters, three nodes per cluster, ~300GB/node), ScyllaDB delivered about 4× the throughput and 7× lower response times with far more stable latency. Maintenance tasks were much faster (repairs ~2× faster; decommissions cut from ~3 hours to ~0.5 hour), and Allegro found it could replace three Cassandra nodes with one ScyllaDB node—reducing hardware costs while improving responsiveness—so the company is migrating from Cassandra to ScyllaDB.

Allegro

Lukasz Pachciarek

Senior Database Administrator