MCRL2: Multi-resource Cross-attention-based Representation Learning-augmented Reinforcement Learning for Cloud Microservice Scheduling

MCRL2 is a reinforcement learning framework that enhances microservice scheduling in cloud data centers by integrating multi-resource cross-attention-based representation learning. It introduces MCRL, a representation learning component that captures structured interactions among nodes, resources, and microservices, and couples this with an actor‑critic architecture and a maximum entropy objective. Experiments on real production cluster traces show that MCRL2 outperforms existing baselines in load balancing, scheduling success rate, and average completion time across diverse workloads.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

• reinforcement-learning