Cassandra Performance Tuning Like You've Been Doing It for Ten Years
summary
Explains a repeatable observe‑orient‑decide‑act methodology plus Brendan Gregg’s USE checks (utilization, saturation, errors) and concrete tools: iostat/mpstat, eBPF/BCC (cachetop, biolatency, ext4dist), and flame graphs. Shows a measured fix: lower compression chunk to 4KB for tiny partitions to cut read amplification, disables read‑ahead to get up to 10x improvements on counters/LWT, and uses TLP stress/TLP cluster for lab benchmarks; changes prove effective in multi‑hundred‑node fleets.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 5:20Performance tuning matters
- 11:21Problem-solving methodologies
- 22:17Read amplification and compression
- 23:39Disabling read-ahead
- 25:02Practice and benchmarking
- 28:58Training opportunities
- 30:54Q&A
related talks
17mToo Many Knobs to Tune? Towards Faster Database Tuning by Pre-selecting Important
39mMaking Cassandra Faster in Cloud-native architecture
23mDeciphering Cassandra’s Prophecy: Preventing the Fall of Your Application
40mLatency, Throughput & Fault Tolerance: Building a Modern Database
45mCassandra Architecture: Understanding, Scaling and Optimizing
44mDemystifying Benchmarks: How to Use Them To Better Evaluate Databases
01h 00m7 Tips to Performance Tuning, Optimization, and Everything
45m8ms/99th write percentile latency - is it fast? (...)