Abstract
The proliferation of event-driven microservices and the strategic adoption of multi-cloud environments present significant challenges for efficient resource provisioning. Traditional static or rule-based autoscaling mechanisms often fall short in adapting to highly dynamic and unpredictable workloads characteristic of event-driven architectures, leading to either over-provisioning (increased cost) or under-provisioning (performance degradation). This paper proposes a novel dynamic resource provisioning model specifically designed for multi-cloud deployments of event-driven microservices, leveraging Reinforcement Learning (RL) coupled with real-time workload analysis. Our model continuously monitors key performance indicators and workload patterns across heterogeneous cloud providers, allowing an intelligent RL agent to make autonomous scaling and placement decisions. By training the agent with a reward function optimized for both performance (e.g., latency, throughput) and cost efficiency, the system dynamically allocates and deallocates resources, and orchestrates microservice instances across multiple clouds. Experimental results demonstrate that our proposed model significantly improves resource utilization, reduces operational costs, and enhances application responsiveness compared to conventional autoscaling strategies, offering a robust and adaptive solution for complex distributed systems.
