SE-DO Framework

The rapid growth of interconnected IoT devices has introduced complexities in their monitoring and management. Autonomous and intelligent management systems are essential for addressing these challenges and achieving self-healing, self-configuring, and self-managing networks. Intelligent agents have emerged as a powerful solution for autonomous network design, but their dynamic and intelligent management requires processing large volumes of data for training network function agents. This poses significant challenges for resource-constrained environments like IoT devices, which have limited computational power, network bandwidth, and power consumption capabilities.In this paper, we propose a scalable and comprehensive approach called Scalable and Efficient DevOps (SE-DO) to optimize the performance of intelligent agents in resource-constrained environments using a multi-agent system architecture. Our approach leverages a multi-agent-based service design that enables both reactive responses and proactive anticipation and reconfiguration of the network system to meet dynamic requirements. This approach is particularly suitable for next-generation networks like 6G, which demand highly efficient and reliable solutions to support emerging services and applications.To demonstrate the effectiveness of our approach, we implement a multi-agent system comprising a data collector agent, a data analytics/preprocessing agent, a data training agent, and a data predictor agent. We analyze the impact of different machine learning models, including ANN, CNN, and RNN, on each agent's performance while considering resource constraints in both micro-service and agent-based approaches. Through experiments on real-world data, our proposed architecture achieves high accuracy and efficiency within the limitations of resource-constrained environments

随着物联网设备的互联互通迅速增长，其监控与管理呈现出复杂性。针对这些挑战，实现自我修复、自我配置和自我管理的网络，自主和智能管理系统变得不可或缺。智能代理作为自主网络设计的强大解决方案，其动态和智能化管理需要处理大量数据以训练网络功能代理，这为计算能力、网络带宽和能耗能力有限的物联网设备等资源受限环境带来了重大挑战。在本研究中，我们提出了一种名为可扩展与高效DevOps（SE-DO）的全面解决方案，通过多智能体系统架构优化资源受限环境中智能代理的性能。我们的方法利用基于多智能体的服务设计，既能够实现网络系统的反应式响应，又能够进行主动的预测和重构，以满足动态需求。该方法特别适用于下一代网络，如6G，它需要高度高效和可靠的解决方案来支持新兴的服务和应用。为了验证我们方法的有效性，我们实施了一个包含数据收集代理、数据分析/预处理代理、数据训练代理和数据预测代理的多智能体系统。在考虑微服务和基于代理方法中的资源限制的同时，我们分析了包括人工神经网络（ANN）、卷积神经网络（CNN）和循环神经网络（RNN）在内的不同机器学习模型对每个代理性能的影响。通过在真实世界数据上的实验，我们提出的架构在资源受限环境的限制下实现了高精度和效率。