Netty 5: Lessons Learned
summary
Netty 5 removes costly handler traversal with a new skip annotation that reduces per-read call sites and, in a five-handler microbenchmark, raises operations from ~230k/s to ~750k/s (~3x throughput). The pipeline now caches a bitmask in a thread-local cache to avoid repeated reflection, and the event loop splits selection from task execution to expose metrics and extensibility. A pooled direct memory allocator plus pooled ByteBuf reuse collapses allocation churn to ~1k objects in steady state. Roadmap targets 5.0 alpha in 2020; profiling uses async‑profiler/YourKit.
AI-written summary · about
chapters (10)
- 0:00Introduction and Netty overview
- 4:31Why Netty 5 was needed
- 5:58Stack traces and channel pipelines
- 13:39Event loop extensibility
- 21:09Buffer pooling and memory issues
- 29:55Netty 5 roadmap and features
- 32:49Java 8+ features in Netty 5
- 34:54Performance bottleneck discovery
- 36:44HTTP/2 rework plans
- 38:29Q&A and closing
related talks
45mNetty, Meet THE IO Framework for the JVM
41mPerformance Story: How to Speed Up HTTP/2 Client
39mAmazing WebSockets in Netty
23mBuilding a network stack for optimal throughput / low-latency trade-offs
46mReactive ❤️ Loom: A Forbidden Love Story
38mReactive ❤️ Loom: A Forbidden Love Story
52mMicro-Optimization for Fun and Profit
51mThe Need for Async