Playing Games In The Clouds
summary
Two example mechanisms and concrete outcomes: with two nodes of capacity 5 and 2 handling 4 jobs the Nash-style allocation sends 3.5 jobs to the faster node, yielding a max response ~1.33s; second-price auction incentives make truthful capacity bids a dominant strategy so nodes report real memory and the scheduler can pick the best host; Cloud Foundry uses auction-like placement; an allocation algorithm that drops nodes with negative assigned jobs (too slow) reduces overload and cuts variance compared with proportional or naive schemes.
AI-written summary · about
chapters (10)
- 0:00Start
- 3:08WHAT IS A GAME?
- 7:34THE PROBLEM
- 10:20A NASH BARGAINING SOLUTION
- 16:01THE SOLUTION
- 18:09OPTIMAL JOB PLACEMENT
- 19:10ALLOCATION ALGORITHM
- 20:55APP PLACEMENT IN CLOUD FOUNDRY
- 23:59LEVELLING UP
- 24:55A WARM UP
related talks
05mSponsored Keynote: The Cloud (And Your $$$) is Not Infinite: Dynamic Scheduling at Every Layer
55mCloud services for gaming
10mLightning Talk: Resource Fairness and Utilization for Heterogeneous Batch/M...
05mResource Roulette: Winning the Kubernetes Allocation Game
29mScale Your Batch / Big Data / AI Workloads Beyond the K8s...
19mKeynote: Building Bridges: Cloud Native and High Performance Computing
21mExperience Operating Large GPU Clusters at Organizational Scale
28mGPU Reservations: Maximizing Utilization and Fairness Across Teams