Benchmark testing
summary
Compares real benchmark outcomes rather than theory: 24 vs 12 drives flips when measured as performance per dollar, and Gluster dispersed vs replicated volumes change the winner depending on client protocol. Uses fio, smallfile, iozone, spec SFS and DVDStore2/3 for full‑stack VM density tests; monitoring with prometheus + grafana or collectd/graphite; wrap tests with pbench or scripts. Finds plateaus — total OPM tops near ~7,500 (per‑stack OPM falls), suggests ~45 VMs per cluster for headroom, run long iterations, correlate metrics, automate.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:35Why performance tests
- 9:25How to test
- 14:56Tools
- 21:38Testing tools
- 26:00Controller nodes
- 26:39Wrapper tools
- 27:58Workload generators
- 29:50The next step
- 31:30The graph
- 36:33Automating
- 37:15End questions
related talks
40mBenchmarking: Repeatable & Comparable
25mPrescriptively Benchmarking Kubernetes System and Application Using K-Bench
31mTests Are What You Want, What You Really Really Want
20mKubeVirt Benchmar...
06mIGNITE: How to benchmark … poorly
01h 08mPerformance testing in the cloud
40mPerformance Testing and Big Data
49mPerformance Testing