Cloud microservice scheduling improves with cross attention learning
MCRL2: Multi-resource Cross-attention-based Representation Learning-augmented Reinforcement Learning for Cloud Microservice Scheduling
Machine Learning
Summary
Cloud data centers need smart ways to decide which computer handles different small services to keep things running smoothly. This scheduling is tricky because resources vary over time and services use different kinds of resources. The authors created MCRL2, which uses a special learning method called cross-attention to better understand how services, resources, and nodes interact. This improves scheduling decisions, making load balancing more even and tasks finish faster in real cloud workloads.
What this means in practice
- •For cloud infrastructure operators: Make cloud resource allocation more balanced and efficient under changing workloads using advanced learning methods for microservice scheduling.
- •For distributed system engineers: Improve scheduling algorithms by integrating multi-resource awareness and cross-attention representations to handle resource coupling and heterogeneity.
Authors
Tiangang Li, Shi Ying, Xiangbo Tian, Chuan Shi, Ding Xiao
Abstract
Efficient microservice scheduling is crucial for maintaining load balance across nodes in data centers and ensuring high quality of service. However, achieving this in practice remains challenging due to dynamic resource imbalance under fluctuating workloads, nonlinear coupling across multiple resource dimensions, and the heterogeneity of microservice resource demands. While reinforcement learning-based approaches have shown promise, they struggle to capture the complex interdependencies among heterogeneous resources and neglect the importance of learning informative system representations. To address these limitations, we propose MCRL2, a novel reinforcement learning approach augmented with multi-resource cross-attention-based representation learning for microservice scheduling. Specifically, we first propose MCRL, a novel representation learning approach that captures structured and informative interactions among nodes, resources, and microservices via a multi-resource cross-attention mechanism. Then, MCRL2 augments reinforcement learning through MCRL-enhanced actor-critic architecture combined with a maximum entropy objective, improving system state expressiveness and leading to more stable and effective scheduling decisions. Extensive experiments on real production cluster traces demonstrate that MCRL2 significantly outperforms existing baselines in load balancing, scheduling success rate and average completion time across diverse workload patterns.