Planning for scale: How Materialize uses
summary
Materialize moves its metadata and consensus layer onto CockroachDB to get strong consistency with single-digit millisecond latencies. Cockroach outperforms DynamoDB (10–20 ms reads) and avoids RDS/Aurora connection caps, running a 6-node cluster (8 vCPU, 32 GB) that sustains ~2,000 outstanding connections and ~2,000 inserts/sec. Consensus uses a shard_id + auto-increment sequence number table and a tuned SQL pattern to hit Cockroach’s one-phase commit fast path (two network trips); GC interval set to 10 minutes to keep reads fast.
AI-written summary · about
chapters (12)
- 0:00Welcome/Agenda
- 0:52What is Materialize?
- 1:35What are the key properties of Materialize?
- 1:57Transitioning to a cloud-native product
- 3:02Planning and development
- 3:40CockroachDB vs FoundationDB, Amazon RDS, Aurora, Dynamo DB for metadata layer
- 5:36What does Materialize use CockroachDB for?
- 8:01Materialize's high-level cloud architecture
- 8:48Design partnership with Cockroach Labs
- 9:30Going into production and what the future looks like
- 11:01How to optimize performance on CockroachDB
- 13:01Final thoughts
related talks
48mBuilding mission-critical applications with CockroachDB
22mBuilding CockroachDB for performance at scale
15mScaling to support America’s #1 sportsbook
15mScaling to support America’s #1 sportsbook
31mFuture-proofing operational data at scale
42mThe hows and whys of a distributed SQL database
01h 11mFearless global transactions with
52mCockroachDB