Make Descheduler Smarter and Safer: How We Apply Reinforcement Learning in Descheduling Strategies
Talkat Cloud Native AI Day Europe 2024 · on Tuesday, 2024-03-19· at 14:40· Pavilion 7 | Level 7.3 | Room S03·
summary
Describes applying Reinforcement Learning to Kubernetes descheduling: defining reward functions and training RL agents to handle resource fragmentation and workload hotspots instead of rule-based strategies. Explains using historical scheduling data and a scheduler simulator to train and validate the RL-based descheduler safely without impacting production.
AI-written summary · about
related talks
19mDynaRL: Flexible and Dynamic Scheduling of Large-Scale Reinforcement Learning Training
22mAWARE: Automate Workload Autoscaling with Reinforcement Learning in Production...
25mIntelligent Kubernetes Workload Optimization
17mScaling Reinforcement Learning
10mFair Scheduling for Deep Learning Workloads in Kubernetes
40mLimits, Requests, QoS... Rethinking What You Thought You Knew About Scheduling in Kubernetes
17mReinforcement Learning with Human Feedback in Distributed Systems
29mKubernetes Workload Resiliency in Action: Beyond Basics