Mgmt Config: The future of your autonomous datacenter
summary
Mgmt is a real-time automation engine plus a reactive DSL that pushes streams (date_time, system load, microphone) through resources to act in milliseconds. Demos show hot-plug CPUs into VMs, a scheduler that picks max=2 hosts with TTL=10 using a raft-based cluster, and autoscaling with hysteresis (scale down at load>1.5, wait 10s below before scaling up). Code and examples are on GitHub and the project targets a 0.1 release as production ready.
AI-written summary · about
chapters (16)
- 0:00Start
- 3:30Engine: Design triad
- 7:04Demos: Reactive
- 8:45Demos: Import/Modules
- 9:02Demos: Running the lang
- 13:53Magic CPUs: demo
- 15:06Distributed scheduling: demo
- 16:34Distributed scheduling: docker!
- 20:17Demos: History
- 21:53Hysteresis
- 25:54Future work
- 27:33How can you help?
- 28:10IRC
- 28:44Learn more: Advanced talk tomorrow!
- 28:58Feedback
- 29:37Free Stickers!
related talks
27mKeynote Panel: Future of Cluster Management
15mEvolving Cluster Management
20mGlobal Capacity Management
18mAvoiding and Breaking Out of Capacity Prison
30mBuilding distributed systems is hard - case study (...)
39mPractical Container Scheduling: Juggling Optimizations, Guarantees, and Trade-Offs at Netflix
23mHow cluster managers affect the landscape of modern distributed computing
29mDesign of a Composable Infrastructure Platform