云应用协同自适应优化问题的分散求解方法研究

Constructing cloud application with self-adaptive ability is an important research direction in the field of cloud computing and satisfying the cloud application performance requirement with minimal cost has the practical significance. So, we will take a crack at the coordinating self-adaptive optimizing problem of cloud applications through a distributed collaborative way. Starting from analyzing the factors affecting the application’s load changing, the deep belief net is used to predict the application load jointly. Then, the ways of building the service performance model based on collaborative Filtering and support vector regression, and generating the candidate logic service set based on the tag tree are explored. We will focus on the research of modeling the decentralized optimization decision problem in light of the distributed constraint reasoning theory. The way to solve it by combining the agents negotiating based on distributed constraint optimization and the internal optimal decision based on meta-heuristic will be research thoroughly, to provide the quality guarantee and agent privacy protection property. To tackle the uncertainty of service cost, the auction theory is incorporated into the solving process, and the filtering and compressing algorithm for the candidate service set is also discussed to further improve the constraint solving effect. Finally, we will explore the establishment of a suit of effective validating and evaluating ways for the decentralized solving method.

构建具备自适应能力的云应用是云计算领域的一个重要研究方向，而以最小的成本保证云应用的性能要求具有现实意义。为此将采用分布式协商的方式对云应用协同自适应优化问题的求解方法展开研究。从分析影响应用负载变化的因素入手，研究基于深度置信网的应用负载联合预测方法。进而研究基于联合过滤和支持向量回归相结合的服务性能模型构建及在线调整方法，以及基于标记树的逻辑服务集生成方法。着力研究建立基于分布式约束推理的分散优化决策模型，重点探讨基于分布式约束优化的代理间协商和基于元启发式的内部最优化决策相结合的能够在提供质量保证的同时有效维护各应用隐私的分布式约束求解方法，并研究在求解过程引入拍卖理论以有效应对服务成本的不确定性。同时研究候选服务集的过滤算法及候选服务集的压缩方法以进一步提高约束求解效果，并针对提出的分散求解方法探索建立一套有效的验证及评价方法。

构建具备自适应能力的云应用是云计算领域的一个重要研究方向，而以最小的成本保证云应用的性能要求具有现实意义。为此本课题从分析影响应用负载变化的因素入手，研究了基于深度置信网的应用负载联合预测方法，提出并实现了基于联合过滤和支持向量回归相结合的服务性能模型构建及在线调整方法，以及基于标记树的逻辑服务集生成方法，进而建立了基于分布式约束推理的分散优化决策模型，实现了基于分布式约束优化的代理间协商和基于元启发式的内部最优化决策相结合的能够在提供质量保证的同时有效维护各应用隐私的分布式约束求解方法，在求解过程引入拍卖理论以有效应对服务成本的不确定性，并针对提出的分散求解方法探索建立一套有效的验证及评价方法。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33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。可见，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。