Case Study: Dstillery achieves near-zero failure rates and 40% fewer servers with ScyllaDB

Dstillery Customizes Adtech Audiences at Scale with Scylla

Dstillery, a leading applied data science company for marketing and advertising, relied on daily-refreshed device data and needed to process hundreds of billions of read/write requests per day with latencies under 25 ms. The team hit a scalability and reliability barrier with Cassandra—sustaining a ~0.1% failure rate (thousands of failures) even at low traffic—so Dstillery sought an alternative database and selected ScyllaDB.

Dstillery migrated its performance-sensitive workloads to ScyllaDB, benefiting from ScyllaDB’s self-tuning architecture and lower operational overhead. The move drove failures down to nearly 0.0% (hundreds vs. thousands), with an extreme-case drop from 14% to 0.8%, reduced one cluster’s server count by about 40%, and now runs eight ScyllaDB clusters (~130 nodes each across two DCs) to store device location history and attributes—improving reliability, reducing costs, and supporting higher-quality audience modeling.

Dstillery

Hang Chan

Site Reliability Engineer