Partisan: high performance large scale distributed Erlang

on YouTube ↗46m

" summary

Partisan is a user-space alternative Erlang runtime that lets applications pick network overlays at runtime (full-mesh, client‑server, peer‑to‑peer) and ships named channels and affinitized scheduling plus multiple TCP connections to reduce head‑of‑line blocking. It adds a deterministic replay and fault‑injection framework, plugs into Kubernetes/Mesos/Docker, and finds real bugs in distributed systems like React. Benchmarks show order‑of‑magnitude wins—up to 34x improvement on high‑latency large‑payload workloads—and client‑server collapses at ~512 nodes without overlay specialization.

AI-written summary · about

" chapters (10)

1. 0:00Introduction to Actors
2. 2:58Limitations of Distributed Erlang
3. 5:31Introducing Partisan
4. 8:46Network Topology Strategies
5. 13:23Latency Reduction Techniques
6. 17:16Performance Results
7. 24:23Partisan Features
8. 28:55Message Interposition
9. 31:44Resilience Verification
10. 39:59Conclusion and Q&A

" related talks

• 22mPARTISAN: Scaling the Distributed Actor RuntimeChristopher S. Meiklejohn, Heather Miller, Peter Alvaro · USENIX ATC '19
• 32mGetting distributed with FirenestMichał Muskała · Code BEAM Lite Berlin 2018
• 33mDiscovering distributed systems in ElixirMaciej Kaszubowski · Code Elixir LDN 2018
• 45mBuilding Available and Partition Tolerant...Gabi Zuniga · ElixirConf 2016
• 31mAn Open Erlang PaaS Done RightVladimir Kirillov · Erlang Factory SF Bay Area 2014
• 33mA Reliability Benchmark for Actor-Based Server LanguagesPhil Trinder, Aidan Randtoul · Erlang 2022
• 36mKnee Deep Into P2PFernando Mendes · Lambda Days 2018
• 39mAn adventure in distributed programmingWiebe Marten Wijnja · ElixirConf EU 2019