Avoiding and Breaking Out of Capacity Prison
summary
Describes a working capacity-management system used on Microsoft Azure Storage that automates cluster rebalancing, forecasting, and playbooked mitigations. Time-series dashboards, threshold and trend alerts, and forecast models drive decisions; unexpected supply shocks and multi-hour migrations are handled by tested levers like throttling marketing or automated data moves. The team implements vector bin packing to place thousands of objects across 7 dimensions, rebalancing clusters from hotspots (98%) into a target range 55%, cutting costs and avoiding immediate hardware purchases.
AI-written summary · about
chapters (12)
- 0:00Intro
- 1:00Identify
- 1:31Monitor f you're not monitoring capacity, you're not monitoring your service.
- 2:28Efficiency
- 3:43Forecast
- 5:59Getting Hardware
- 9:07Playbook
- 9:34Virtual Team
- 10:56Building Levers
- 14:14Challenges
- 14:502-Dimensional Balancing
- 15:55Legal Movement
related talks
16mGlobal Capacity Management With Flux
20mGlobal Capacity Management
20mGlobal Capacity Management with Flux
18mReal-time adaptive controls for resilient container workloads
30mUnderstanding Bottlenecks & Conquering them
30mMgmt Config: The future of your autonomous datacenter
21mHandling Traffic Spikes Through Automated Prioritized Load Shedding
32mSystems @Scale Summer Live Q&A Session #4