Why Data Access Patterns Have Always Mattered
summary
Argues that memory layout and access order still drive performance, backing it with a 6502 assembly comparison where a linked list costs 49–64 cycles vs 43–57 cycles per iteration and a HDD/SSD benchmark showing random access up to 100x slower than sequential. Presents historical data showing RAM cost falling from $24,000 to about $0.11 per MB, traces why reliance on caches after ~2005 made locality lessons fade, and demonstrates compact, linear storage (spans/arrays) measurably outperforms pointer-chasing lists.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:01Lists versus vectors
- 6:49Cache complexity
- 11:11Storage and disk mechanics
- 18:26Disk optimization and layout
- 21:44Data access patterns
- 25:35Modern caching reality
- 29:06Assembly performance analysis
- 34:28Historical timeline of computing
- 42:46Hypothesis on modern performance
- 45:49Audience discussion and reflection
- 51:27Final conclusions and takeaway
related talks
42mAdventures in Efficiency and Performance
51mDesigning for Performance
01h 04mWe Live in Interesting Times
50mLet's talk about the cache!
06mEarly experience of Utilizing Persistent Memory for Database... (Work-in-Progress Report)
48mDesign of data-intensive applications
45mBeyond Lists: High Performance Data Structures
16mJohnny Cache: the End of DRAM Cache Conflicts (in Tiered Main Memory Systems)