High Performance Network Programming in Rust
summary
The Zenoh stack is rewritten in Rust and achieves near-iperf throughput on 100 Gb/s networks with about 50 Gb/s at 60 KB and 40 Gb/s at 8 KB, saturates 10 Gb/s with 512 B and 1 Gb/s with 32 B, and measures about 30 microsecond latency end-to-end. Zenoh outperforms MQTT roughly fourfold in messages/sec, runs peer-to-peer or brokered, and avoids TCP when useful. Engineering lessons: prefer async with limited threads, minimize dynamic allocations in critical paths, watch future state size, and measure runtimes — tokio shows higher latency under contention compared to async-std for these workloads.
AI-written summary · about
chapters (19)
- 0:00Intro
- 1:47How High is High Performance?
- 2:51Scalability
- 7:47Runs Everywhere
- 10:41Abstractions
- 12:10Putting and Querying Data
- 14:47Any Topology
- 16:26Extensible by Design
- 18:56Throughput in Perspective
- 25:27Productivity
- 26:40Asynchronous Programming
- 28:35R Rust
- 31:08Concurrency
- 35:56Memory
- 38:19Measure don't Guess
- 38:55Key Libraries
- 39:25Popularity vs Performance
- 40:26Impact of Async Tasks
- 42:54Step after Step
related talks
41mEnergy Consumption and Communicating Protocols
34mMove fast and don't break things: High-performance networking in Rust
36mEclipse Zenoh
54mzenoh: A Next-Generation Protocol for IoT and Edge Computing
31mThe Zenoh Protocol: Zero Overhead Pub/sub, Store/Query and Compute
56mThe RUST Language for High‑Load Network Services — Quick Start
24mHow We Built Protosockets to Go Beyond HTTP and gRPC
23mFast and efficient network protocols in Rust