The Joy of Building Large Scale Systems
summary
Argues modern hardware is underused and gives concrete wins: switching Java to ZGC produces submillisecond GC pauses, pinning network IRQs cuts tail latency by 71% reduction in tail latencies, moving services to AWS Graviton yields ~30% price to performance improvement, and adopting kernel async IO with io_uring unlocks massive random I/O throughput. Shows a Rust ISO8601 extension that gives a 10x date parsing speedup, a JMX exporter fix that halves Kafka CPU, and mmap/fileformat tweaks that cut model load times 100x.
AI-written summary · about
chapters (15)
- 0:00Intro
- 1:57Wait, but why?
- 6:10Latency numbers
- 6:35B-Trees
- 10:37Hardware
- 13:00CPUs
- 15:47Networks
- 18:48Free lunch?
- 22:10Thread per core
- 27:52Systems programing languages
- 29:43Date parsing
- 33:21New tricks
- 38:47eBPF
- 42:55Wait, but why?
- 44:44Outro
