Causal's 1000x Spreadsheet; Performance beyond profiling
summary
Engine moved from browser to Node then rewrote in Go, using go tool pprof, causal profiling (cause), sync.Pool and memoization to find high‑ROI fixes. Napkin math shows two billion‑cell operations hit memory bandwidth limits (8 GB per large variable, baseline ~700 ms), which drives a shift from hashmaps to struct‑of‑arrays and cache‑line packing. Microbenchmarks cut runtime from 6s to 0.5s, production reaches 150M cells, and CPU family changes (Ice Lake) yield ~30–50% speedups.
AI-written summary · about
chapters (4)
- 0:00Start
- 8:42Profiling for fun and profit
- 11:10Causal profiling concurrent programs with coz
- 13:08Practice safe concurrency
related talks
01h 21mPerformance Really Matters
56mWrite Fast Code Like a Native
01h 04mUnlocking Performance Through Reverse Engineering
01h 50mUnlocking Modern CPU Power - Next-Gen C++ Optimization Techniques
34mFast Code Without Breaking Things
27mHigh Performance Architecture
51mDesigning for Performance
45mThe Joy of Building Large Scale Systems