Benchmark Your Cloud Native Database
summary
Benchmarks compare bare metal, node-local volumes, and Rook/Ceph on a six‑blade cluster (20 cores, 128GB RAM, two 200GB SSDs) running sysbench/fio, pgbench and CockroachDB TPCC/bank microtests with mostly in‑memory datasets. Results show local PV ≈ bare metal for throughput and latency, while Rook/Ceph introduces about 2× latency and roughly half the DB throughput (sequential reads can be faster via multi‑replica fetch). Shared networks and cloud noisy neighbors mask differences, so choose cloud‑native storage only if redundancy/management tradeoffs justify higher latency.
AI-written summary · about
chapters (21)
- 0:00Introduction
- 1:16Why run a database on Kubernetes
- 3:07Why performance is critical
- 5:23Types of storage
- 7:37Latency and throughput
- 8:46Micro benchmarks
- 10:40Cobbler benchmarks
- 11:19Benchmarking tips
- 12:57Why use bare metal
- 13:58Disclaimer
- 15:02Bare Metal Platform
- 16:40First set of tests
- 17:15SSD throughput
- 18:07Database benchmarks
- 19:55Use local volumes
- 21:46Shared network
- 22:43Storage configuration
- 23:56Results
- 26:24Whats next
- 26:51Conclusions
- 27:31Questions
related talks
32mBenchmark Your Cloud Native Database
39mBenchmarking Cloud Native Storage
05mIgnite - How Fast Was My Kubernetes?
56mBenchmark Your Cloud-Native Database
33mBenchmarking Cloud Native Databases Performance on Kubernetes
47mThe Road Towards an Open Database Performance Ranking
17mComparing DBs with Golang in the cloud
36mBest Practices in PostgreSQL Tuning: Navigating Key Performance Bottlenecks in the Cloud