Deft Scheduling of Dynamic Cloud Workflows with Varying Deadlines via Mixture-of-Experts
Researchers have developed DEFT, a novel deep reinforcement learning scheduler for dynamic cloud workflows. DEFT utilizes a mixture-of-experts architecture, with each expert specialized for different deadline tightness levels. This approach allows for adaptive routing of decisions through the most appropriate experts, enabling DEFT to meet a wider range of deadline requirements than single-expert systems. Experiments show DEFT significantly reduces execution costs and deadline violations compared to existing methods. AI
IMPACT Introduces a novel scheduling approach that could improve efficiency and reduce costs in cloud computing environments.