Ray: A distributed system for emerging AI applications
summary
Ray is an open source distributed system that implements task and actor abstractions for Python ML workloads, with a C++ backend, Python workers, an in‑memory shared object store (Apache Arrow) and a Redis‑based global control store. It claims fault tolerance by forwarding uncommitted lineage stashes between schedulers so recovery replays only lost tasks, keeping millisecond task latencies even at ~2,000 tasks/sec per node; used at Ant Financial for fraud detection.
AI-written summary · about
chapters (12)
- 0:00Start
- 2:05The Machine Learning Ecosystem
- 4:18What is Ray?
- 7:24A growing number of production use cases
- 8:24Ray API
- 12:33Parameter Server Example
- 18:23A scalable architecture for high-throughput. fine-grained tasks
- 22:39Fault tolerance: Lineage reconstruction
- 26:54Previous solutions committing first for correctness
- 31:39Lineage stash: Fault tolerance for free
- 33:38Conclusion
- 35:54Lineage stash Rayli commit later
related talks
24mRay: A System for High-performance, Distributed ML Applications
03mRay: A Distributed Execution Framework for Emerging AI Applications
26mRay: A System for High-performance, Distributed Python Applications
22mRay, a Unified Distributed Framework for the Modern AI Stack
29mRay:A Distributed Computing Platform for Reinforcement Learning
15mRay: A System for Distributed Applications
01h 31mIntroduction to Ray for distributed and machine learning applications in Python
11mKeynote Q&A