The Quick Journey to Using Ray: How We Implement Ray and Anyscale to Speed up our ML Processes
summary
Anastasia migrates a demand-forecasting pipeline from AWS Batch to Ray with Anyscale, reworking code to queues and actors so every node handles every model. A benchmark job of 100,000 time series (120 points each) on 384 CPU cores shows Ray + Anyscale runs 9x faster and 87% cheaper than the isolated AWS Batch setup. Per-core throughput exceeds Python multiprocessing, spot instances cut costs, and operational work shifts to autoscaler/Anyscale with new devops and sharing challenges.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:27Problem with demand prediction
- 4:17Initial AWS Batch setup
- 7:57Challenges with horizontal scaling
- 9:22Why we chose Ray
- 10:05Ray implementation details
- 12:39Results with Ray
- 14:02Performance and cost comparison
- 15:53New challenges with Ray
- 17:35Why we use Anyscale
- 19:03Conclusion
related talks
32mAnyscale's Ray Data: Revolutionizing Batch Inference
30mScaling Machine Learning at Discord: Our Journey with Ray and Anyscale
30mHow to Scale Unstructured Data Processing with Ray Data
16mFrom Self-Managed Ray to Anyscale Services: Migrating ML Serving
29mScaling ML Pipelines for Autonomous Driving with Ray
24mBuilding Complex Data Analytics Pipelines with Ray
15mRay: A System for Distributed Applications
35mOperating a Production-Scale Physical AI Data Platform with Ray