When every Microseconds counts: Lessons learned about performance
summary
Claims: measurement beats assumption — a custom pre-allocated memory pool sometimes loses to malloc until hot-segment reuse is fixed, and fixing it yields a clear latency win. Implements a performance graph of per-action latency to baseline changes (too costly for continuous production, but invaluable for debugging). Shows that reducing branching and moving decisions to compile/init time improves predictability, and that an err-on-fast-path design trades safety for deterministic microsecond gains but risks hard-to-debug failures; rare flows can dominate latency.
AI-written summary · about
chapters (7)
- 0:00Intro
- 0:16Performance
- 0:57Its not faster if it was measured
- 2:23Branching is expensive
- 3:03Safety check
- 3:31The price
- 4:22The critical flow
related talks
01h 09mComputer Architecture, C++, and High Performance
32mThe Impact of Memory Allocators on Performance: A Deep Dive
55mThe performance Addict's Toolbox
01h 14mEfficiency with Algorithms, Performance with Data Structures
01h 04mCPU design effects
04mZero-Overhead Compiler Pessimization
01h 05mThe Speed Game: Automated Trading Systems in C++
01h 00mIntroduction to Hardware Efficiency in Cpp