WASM in ByteFaaS, Past, Present, and Future
Talkat WasmCon 2023 · on Thursday, 2023-09-07· at 11:00· Grand Ballroom E, Hyatt Regency Bellevue, 900 Bellevue Way NE·
summary
ByteFaaS replaces containers for many short-lived functions with a Wasm worker running Wasmtime + Tokio, precompiles user modules, and adds a service-provider hostcall layer (KV, discovery, HTTP, MQ, io_uring) so Wasm functions call backend services without full containers. Co‑locating called functions cuts remote-call average latency from ~10–20 ms to ~2–3 ms; measured cold starts are cold start <10ms and cached activations cached start <5ms. Ongoing work adds Ray scheduling for distributed Wasm tasks and Llama 2 via llama2.c inference.
AI-written summary · about
chapters (10)
- 0:00WebAssembly in ByteFaaS intro
- 1:02Legacy ByteFaaS limits
- 6:30Lightweight WASM worker design
- 11:31Optimizing WASM deployments
- 15:45Next-gen ByteFaaS with Service Weaver
- 20:50WASM Ray architecture
- 24:05Live demo: WASM Ray integration
- 27:06WASM Match inference demo
- 29:19Future work and features
- 31:10Q&A
related talks
12mFaaS Platform Engineering with Wasm
22mPlatform Engineers assemble! Create your FaaS Platform using WASM
19mWebAssembly's Journey Beyond the Browser
37mBeyond the browser: Moving forward with a more Declarative Future with WebAssembly
43mMeet Your Future - WebAssembly
34mContainers vs. WebAssembly: What's the Difference and Which Should I Use?
31mHow We Can Create a Thriving Wasm Ecosystem
53mWASM Wonders Decoding WebAssembly's Inner Workings