Faster or better designed? Choose any two!
summary
Design shows a ranked set of practical rules: adopt no garbage zero-allocation fast paths, use natural batching with a ring buffer and a dedicated sender thread, enforce the single writer principle to avoid locks, and handle exceptional cases off the hot path. Aeron demos hit ~40 µs median RTT and a 3M msg/s DPDK run (~7.8–8 Gbit/s AWS single-flow cap); Java matches C when both implement the same zero-copy, lock-free techniques.
AI-written summary · about
chapters (24)
- 0:00Introduction
- 0:49What is latency
- 1:35What is low latency
- 3:22Design principles
- 4:12No garbage
- 6:04Simple example
- 6:25Key and value parsing
- 9:34Performance comparison
- 11:21Reducing garbage
- 17:40Threading
- 18:35Single Thread
- 19:30Barly Serial Problem
- 20:12Threads
- 22:04Exceptional logic
- 22:59Fragments
- 26:13Single Writer
- 29:07Sharing
- 31:20Copying
- 34:52Aeron
- 37:26Aeron in the cloud
- 38:19Echo Benchmark
- 38:50Latency Test
- 39:55throughput Test
- 41:22GCP
related talks
13mLive in the Express Lane
31mPatterns of Low Latency
01h 11mAsychronous design with Spring and RTI: 1M events per second
06mRDMA Programming Design and Case Studies - For Better Performance Distributed
28mRock around the clock (synchronization
40mEverything Will Flow
34mDesigning, Building and Testing Deterministic HFT Systems
02h 42m!!! Do not publish or share !!!!