L7-COLO: A Cost-Aware Layer-7 Load Balancing Framework with Deep Traffic Prediction for Dynamic Optimization of Enterprise Cloud Services

Abstract

Modern enterprise cloud applications increasingly rely on microservice architectures and container orchestration platforms, where Layer-7 (L7) load balancing plays a critical role in routing application-level traffic among service instances. Conventional L7 routing strategies—such as round-robin or least-connection—are primarily designed to improve performance and availability, but they rarely consider the impact of routing decisions on cloud resource consumption and operational cost. In large-scale SaaS platforms, e-commerce systems, and financial cloud services, inefficient traffic distribution may trigger unnecessary container scaling, bandwidth overhead, and suboptimal resource utilization, leading to significant increases in cloud service costs. To address this challenge, this paper proposes L7-COLO (Layer-7 Cost-Optimized Load Optimization), a cost-aware intelligent load balancing framework that integrates deep traffic prediction with dynamic routing optimization for enterprise cloud environments. The proposed approach constructs a Traffic–Resource–Cost (TRC) mapping model that captures the relationship between predicted application traffic, infrastructure resource usage, and cloud operational cost. A CNN-BiLSTM-Attention model is first employed to forecast future API request traffic, followed by a resource estimation module that predicts CPU, memory, bandwidth, and container scaling requirements under different routing strategies. Based on the predicted resource usage, a cloud cost model evaluates the expected operational cost, while a reinforcement learning policy optimizer dynamically adjusts L7 routing weights to minimize cost under service-level agreement constraints. Experiments conducted on real-world microservice workload traces demonstrate that the proposed L7-COLO framework significantly improves cloud resource efficiency and cost control. Compared with traditional Round-Robin routing, L7-COLO reduces overall cloud operational cost from 18.9 USD/min to 12.9 USD/min, achieving approximately 31.7% cost reduction, while also improving resource utilization to 78.4%. Meanwhile, the framework maintains stable service performance with an average latency of 155 ms and reduced SLA violation rates, demonstrating its effectiveness for enterprise cloud cost optimization.