面向整体收益优化的多SLA云服务系统动态自适应决策方法

Self-adaptively guaranteeing performance of cloud service while minimizing resource cost is an importance problem in the area of cloud service, and making cloud service with self-adaptive ability is an important development direction for the cloud computing. In this project, the self-adaptive method for the cloud service system with multiple SLAs based on the market theory and decentralized decision making is explored from the global revenue oriented optimization of the cloud service provider and every kinds of SLA user. The self-adaptive decentralized decision mechanism based on bid price theory, the heuristic rules based on market theory and utility constraint decomposition algorithm based on distributed constraint method are researched through constructing the dynamic self-adaptive decentralized decision making with multiple SLAs cloud service system by building the global revenue oriented optimization considering the self-adaptive decentralized decision making model and self-adaptive utility decision making model. To make the data aware and self-adaptive decision more effective, the deep learning is used to model the component service performance. Based on the computed the global revenue oriented self-adaptive action price and earnings, the effectiveness and sustainability of the global self-adaptive decision making is implemented and the optimized component resource adjusting action is formed. In this project, the market model is employed to build the self-adaptive utility model and the decentralized decision making is research based on the distributed constraint solving technique. It will provide new ideas and theoretical basis for the self-adaptive swarm decision making.

自适应地保证云服务性能要求、又能最小化资源成本是云服务领域面临的一个非常重要的问题，具备自适应能力的云服务是云计算的重要发展方向。本课题针对云服务提供商和各类SLA用户的整体收益优化问题，探索基于市场理论的多SLA云服务系统性能优化动态自适应决策方法。通过构建面向系统整体收益的自适应分散决策模型和自适应决策效用模型，研究基于竞价理论的自适应分散决策机制、基于市场理论的启发式规则和基于分布式约束求解的效用约束分解算法，形成多SLA云服务系统动态自适应分散决策的理论体系；通过基于深度学习的组件服务性能建模，实现有效的环境数据感知，支持自适应决策的有效性；通过面向整体收益的自适应动作代价/收益计算，形成符合整体收益优化的组件服务资源调整动作，实现自适应决策的整体有效性和可持续性。本课题基于市场模型的自适应问题效用建模、结合分布式约束求解的分散决策方法，将为群体自适应决策研究提供新思路和理论依据。

自适应地保证云服务性能要求、又能最小化资源成本是云服务领域面临的一个非常重要的问题，具备自适应能力的云服务是云计算的重要发展方向。本课题针对云服务提供商和各类SLA用户的整体收益优化问题，建立了基于市场理论的多SLA云服务系统性能优化动态自适应决策方法。通过构建面向系统整体收益的自适应分散决策模型和自适应决策效用模型，实现了基于竞价理论的自适应分散决策机制、基于市场理论的启发式规则和基于分布式约束求解的效用约束分解算法，形成了多SLA云服务系统动态自适应分散决策的理论体系；通过基于深度学习的组件服务性能建模，实现了有效的环境数据感知以支持自适应决策的有效性；通过面向整体收益的自适应动作代价/收益计算，形成了符合整体收益优化的组件服务资源调整动作，实现了自适应决策的整体有效性和可持续性。本课题基于市场模型的自适应问题效用建模、结合分布式约束求解的分散决策方法，将为群体自适应决策研究提供新思路和理论依据。目前我们已经完成了相应的研究工作，达到了研究目标的要求。研究成果体现在：国内外期刊和会议上共发表学术论文38篇，出版专著2部，培养博士研究生5名，培养硕士研究生10名。开展的研究对推动云计算技术的普及和方法应用具有重要意义。