Keynote Address: Everything You Always Wanted to Know about Storage Analysis...
summary
Thirty years of storage work distills into concrete advice: expect thousands of experiments (a small matrix of micro/macro/trace runs plus varied sizes and repeats easily makes 4,000+ experiments and weeks of wall time), measure device limits first, and isolate hotspots with TSC-based histogram sampling (negligible memory, ~4% CPU overhead historically). Use eBPF/tracepoints carefully, show progressions over time, prefer boxplots or log scales, and share big real data like the 5+ TB dedupe dataset that now has ~250 citations.
AI-written summary · about
chapters (13)
- 0:00Storage analysis introduction
- 2:24Storage system limits
- 3:50Modern storage stack complexity
- 7:21Storage benchmarking tools
- 10:50Presenting benchmarking data
- 21:26Visualizing storage metrics
- 26:02Storage performance bottlenecks
- 33:26Experiment duration importance
- 36:17Storage systems research
- 42:17AI in storage research
- 44:02Scaling storage problems
- 45:47Survey papers value
- 48:16Closing remarks and Q&A
related talks
56m25 Years of Storage Research and Education: A Retrospective
01h 00mMeasure, Then Build
14mFoundational HPC Systems for 2020 and Beyond
45mA Brief History of Data Storage
24mThe Art of Computers
21mThe Resilient Hard Disk Drive (1984 to 2062)
14mKeynote: Instruction Sets Want to be Free - A 10 Year Retrospective
55mComputing in the 1960s