How We Saved Cloud Budget by Using Autoscaling
summary
A case study of a recommendation platform built with state-of-the-art ML algorithms that consumed very large compute (example: ~930 CPUs and 4.7 TB RAM per instance before optimization). It covers cost savings on cloud resources achieved by applying scaling techniques to reduce CPU/RAM usage.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:30What are recommendations
- 3:21Platform and architecture overview
- 5:31Cloud cost growth issues
- 6:35Dynamic environments solution
- 8:50Feature environments explained
- 12:33Certificate and Jenkins issues
- 15:07Dynamic testing environments
- 18:44Kubernetes autoscaling
- 21:16HPA setup and custom metrics
- 24:27Summary and budget savings
- 27:18Recommendations and conclusions
- 28:08Q&A
- 33:59Technology selection questions
- 34:44GPU and quotas questions
- 35:47Testing environment costs questions
- 37:12Environment management questions
- 38:40Environment monitoring questions
- 40:28Closing remarks and awards
related talks
31mAccelerate AI adoption: automated cloud efficiency
36mMigrating Applications to the Cloud in a Cost Effective Manner
24mOptimizing Cloud Costs with Machine Learning - A CloudVane Story
31mAutoscaling Apps without Supersizing Costs
21mBest Practices for Optimizing Your Cloud Spend with AI
45mAutoscale — Pay When You Earn
33mHow we do cloud cost efficiency at Notion
20mHow the cloud changed the way I do analytics and research