Human-Computer Collaboration An Integrated Graph Query Processing Framework Using Deep Reinforcement Learning and Adaptive Indexing in Heterogeneous Cloud Environment

The advent of heterogeneous cloud infrastructures, characterized by diverse hardware accelerators (CPUs,GPUs,FPGAs) and multi-tiered storage systems (NVMe,SSD,HDD),has revolutionized data processing capabilities while introducing formidable challenges for graph query execution. Traditional graph processing systems, designed for homogeneous environments, falter under hardware disparity, leading to severe performance degradation, sub optimal resource utilization, and inflated operational costs. This paper presents HetroGraph-OPT, a novel, integrated framework that synergistically combines Multi-Agent Reinforcement Learning (MADRL) with a Multi-Level Adaptive Index (MLAI)to autonomously optimize graph query processing in such complex, non-uniform settings.By employing a context-aware MADRL agent for intelligent query distribution and real-time resource orchestration, alongside a hardware-conscious MLAI that dynamically adapts its structure to local node capabilities, HetroGraph-OPT achieves unprecedented performance equilibrium. Implemented natively on Kubernetes with seamless multi-cloud support (AWS, Azure, GCP), the framework demonstrates a ۷۲% increase in throughput, a ۵۸% reduction in average query response time, and a ۴۵% improvement in aggregate resource utilization compared to state-of-the-art baselines. Furthermore, it maintains full ACID compliance while scaling efficiently to graphs exceeding one billion vertices, establishing a new benchmark for robust, intelligent, and scalable graph processing in the modern cloud era.

کلیدواژه ها:

Graph Query Processing ، Deep Reinforcement Learning ، Multi-Agent Systems ، Heterogeneous Cloud Computing ، Adaptive Indexing ، Kubernetes Orchestration ، Dynamic Resource Management ، Performance Optimization