基于强化学习的SBS云应用自适应性能优化方法

发布时间：2018-08-20 17:58

【摘要】：自适应的调整云应用所占用的资源是一种有效的保障云应用性能的方法,但传统的决策方法面向基于服务的系统(Service-Based System,SBS)时会存在一些问题,例如基于应用系统性能模型的决策方法不能很好适应云环境下SBS的动态变化,基于智能优化算法的决策方法效率较低.该文提出了一种基于强化学习的SBS云应用自适应性能优化方法.在该方法中,该文建立了自适应基本要素之间相互关系的特征描述框架,利用高层次的系统行为指标(如响应时间、用户并发量、资源量等)来描述系统性能的优化目标等.为了应对云环境以及SBS的动态变化,该文的方法采用了无模型(model-free)的在线学习算法,当用户并发量发生变化导致系统的预期行为发生偏差时,该方法通过不断重复"执行-积累-学习-决策"的过程,可以不断的积累经验数据并优化决策结果.为了保证自适应优化的高效性,该文提出了一种引导算子,可以有效的缩小候选自适应动作的范围,提高算法的学习效率.该文实现了以一个SBS为例的原型框架,使用该框架的实验结果证明了该文提出方法的有效性.
[Abstract]:Adaptive adjustment of the resources consumed by cloud applications is an effective way to ensure the performance of cloud applications, but there are some problems when traditional decision methods are oriented to Service-Based system. For example, the decision method based on the application system performance model can not adapt to the dynamic change of SBS in the cloud environment, and the efficiency of the decision method based on intelligent optimization algorithm is low. This paper presents an adaptive performance optimization method for SBS cloud applications based on reinforcement learning. In this method, a feature description framework of the relationship between adaptive basic elements is established, and high-level system behavior indicators (such as response time, user output, resource, etc.) are used to describe the optimization objectives of system performance. In order to cope with the dynamic changes of cloud environment and SBS, this paper adopts an online learning algorithm without model (model-free). When the user and output changes cause the expected behavior of the system to deviate. By repeating the process of "execution-cumulation-learning-decision", the method can continuously accumulate empirical data and optimize the decision results. In order to ensure the high efficiency of adaptive optimization, this paper proposes a guidance operator, which can effectively reduce the range of candidate adaptive actions and improve the learning efficiency of the algorithm. A prototype framework based on SBS is implemented in this paper. The experimental results show that the proposed method is effective.
【作者单位】：东北大学计算机科学与工程学院;
【基金】：国家自然科学基金(61572117,61300019,61370155) 省科技项目攻关项目(2015302002) 中央高校东北大学基本科研专项基金(N140406002,N150404008)资助~~
【分类号】：TP393.09

【参考文献】