Serverless Computing for Mathematical Optimization
summary
A production pipeline converts a 50‑year battery storage optimization into 600 monthly convex subproblems, applies a convex relaxation into a quadratic program solved via CVXPY, exploits matrices that are ~99.99% sparse to avoid huge memory needs, and runs 600 parallel cloud functions orchestrated by Cloud Composer (Airflow). End‑to‑end runtime drops to about five minutes (a 96% elapsed‑time reduction claimed) at low cost, using open‑source solvers and serverless scaling.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:17Agenda
- 0:50California
- 1:15Renewables
- 1:59The Duck Curve
- 3:23Our Problem
- 5:11Mathematical Optimization
- 5:39When is a problem convex
- 10:03Convex Programming
- 11:10CVX Pi
- 13:09Battery
- 13:40The Point
- 14:20Metrics
- 16:49Serverless Computing
- 18:29Conclusion
- 21:45C vs Rust
related talks
46mServerless for Data Scientists
10mEnergy-Efficient and Cost-Optimized Serverless Computing
45mAutonomous Multi-Cloud Serverless Deployment and Optimized Management
42mServerless Systems That Won't Burn a Hole in Your Pocket
34mBuilding scalable end-to-end deep learning pipelines in the cloud- 42mServerless Systems That Won't Burn a Hole in Your Pocket
40mThe State of Serverless Computing
18mJolteon: Unleashing the Promise of Serverless for Serverless Workflows