Hierarchical Deep RL for Sustainable API Throughput and Latency Optimization in Advertising Clouds

Advertising cloud platforms face escalating challenges in Application Programming Interface (API) performance optimization due to diverse client request patterns, fluctuating workloads, and increasing sustainability requirements. Traditional API management approaches struggle to balance throughput maximization with latency minimization while considering energy efficiency and carbon footprint reduction. The heterogeneous nature of advertising APIs, including bidding interfaces, content delivery services, and analytics endpoints, requires sophisticated optimization strategies that adapt to varying performance requirements and resource constraints. This study proposes a Hierarchical Deep Reinforcement Learning (HDRL) framework for sustainable API throughput and latency optimization in advertising cloud environments. The framework employs a multi-level architecture where global orchestrators manage cross-API resource allocation while local optimizers focus on individual API performance tuning. Deep Q-Networks (DQNs) and Advantage Actor-Critic (A2C) algorithms enable adaptive optimization policies that simultaneously maximize API throughput, minimize response latency, and reduce energy consumption across distributed cloud infrastructure. Experimental evaluation using production advertising cloud workloads demonstrates that the proposed framework achieves 44% improvement in API throughput while reducing average response latency by 39% compared to traditional optimization methods. The sustainability-focused approach decreases energy consumption by 35% and carbon emissions by 42%, while maintaining Service Level Agreement (SLA) compliance rates above 96% across all API categories.