What I learned benchmarking Citus & Postgres performance with HammerDB
summary
Uses HammerDB TPROC-C against Azure Cosmos DB for PostgreSQL with a 1016+32 cluster and moves from 1,000 to 2,000 virtual users to push concurrency; baseline run hits about 1.22M NOPM and after increasing shard_count from 32 to shard_count=40 on 20 workers (352 vCPU total) throughput jumps to 1.75 million NOPM; flipping all_warehouses=false* cuts runaway disk I/O and recovers ~1M NOPM on larger warehouse counts.
AI-written summary · about
chapters (10)
- 0:00Introducing Naisila
- 1:27Naisila’s benchmarking talk begins here
- 4:00OLTP and Citus
- 4:58HammerDB TPROC-C
- 6:28Running benchmarks on Azure Cosmos DB for PostgreSQL
- 9:48Interpreting benchmark results
- 13:30Increasing concurrent connections
- 19:21Adjusting the shard count
- 26:20What I learned about benchmarking
- 28:13Q&A
related talks
52mBuilding Distributed OLTP with Azure Cosmos DB for PostgreSQL
16mOnline Scale-Out and Cloud Data Ingestion
20mLessons learned scaling our SaaS on Postgres to 8+ billion events
02h 02mAzure Cosmos DB: Liftoff
08mFoundations of Distributed PostgreSQL with Citus
13mWhat I learned creating Postgres monitoring metrics for Citus on Azure
28mEfficiently distributing Postgres with Citus – How to choose the right shard key?
52mScaling a Saas application beyond a single Postgres with Citus