Dynamic Leasing of Spare Capacity to Improve Fleet Utilization with Optimus
summary
Optimus is a production platform that leases idle servers to other teams by combining a resource broker, reservation database and allocator with a supply forecasting model that outputs probabilistic capacity guarantees. It supports online ranking (more candidates improves quality), large-scale offline video generation (short bursts coordinated with infra) and delayable async batch jobs, and increases fleet utilization by over 5%. The system exposes a zero SLA pool plus time-shared compute leases and shows reclaim and graceful shutdown procedures that work in practice.
AI-written summary · about
chapters (6)
- 0:00Dynamic leasing introduction
- 7:04Optimus origins in 2013
- 13:29Online ranking success
- 18:06Optimus with probabilistic guarantees
- 22:47COVID-19 leasing surge
- 24:14Future spare capacity directions
related talks
47mScaling FinOps with a holistic approach to resource optimization
30mCapacity Management for Fun & Profit
16mOptimizing Network Provisioning through Cooperation
15mRLBoost: Harvesting Preemptible Cloud Resources for Cost-Efficient Reinforcement Learning
49mProactive optimization for Cloud
31mAccelerate AI adoption: automated cloud efficiency
21mExperience Operating Large GPU Clusters at Organizational Scale
20mThe Coral Reef Cloud: Autoscaling, Reservation, and Preemption