摘要
近年来,基于云计算的软件服务对自适应的资源分配技术提出了越来越高的要求,以保证良好的服务质量(QoS)和合理的资源成本。然而,由于云环境中不断变化的工作负载,基于云计算的软件服务资源分配面临着巨大的挑战,不合理的资源分配方案可能降低QoS,并且导致高额的资源成本。传统的方法大多依赖于专家知识或者多次迭代,这可能导致适应性差和额外的成本。现有的基于强化学习(RL)的方法通常以固定的工作负载环境为目标,不能有效地适应具有可变工作负载的真实场景。为此,提出一种基于深度强化学习(DRL)的自适应资源分配方法,在该方法中根据运行时系统状态训练得到基于深度Q网络(DQN)的管理操作预测模型,并且设计了一种基于反馈控制的运行时决策算法,进而可以根据当前系统状态确定目标资源分配方案。在RUBiS基准对方法进行了评估,实验结果表明,该方法比经典的基于启发式的粒子群优化(PSO)算法和贪心算法适应度函数值平均分别高出4.4%和5.6%,能够有效地平衡对于QoS和资源成本的需求。
In recent years,cloud computing-based software services have put forward increasing requirements on adaptive resource allocation technology to ensure good Quality of Service(QoS)and reasonable resource cost.However,due to the constantly changing workloads in the cloud environment,the resource allocation of cloud computing-based software services is facing huge challenges.The unreasonable resource allocation plan may reduce the QoS and lead to high resource costs.The traditional approaches mostly rely on expert knowledge or numerous iterations,which might lead to weak adaptability and extra costs.Moreover,existing Reinforcement Learning(RL)-based methods target the environment with the fixed workload,and thus they are unable to effectively fit in the real-world scenarios with variable workloads.Therefore,an adaptive resource allocation method based on Deep Reinforcement Learning(DRL)was proposed.In this method,a prediction model of management operation based on Deep Q Network(DQN)was trained according to the runtime system state.A run-time decision algorithm based on feedback control was designed to determine the target resource allocation plan according to the current system state.The method was evaluated on the RUBiS benchmark.The experimental results show that the proposed method has the fitness 4.4%and 5.6%higher than those of the classical heuristic Particle Swarm Optimization(PSO)algorithm and greedy algorithm,respectively,and can effectively balance the demands for QoS and resource cost.
作者
傅德泉
杨立坚
陈哲毅
FU Dequan;YANG Lijian;CHEN Zheyi(College of Computer and Big Data,Fuzhou University,Fuzhou Fujian 350108,China;Department of Computer Science,University of Exeter,Exeter EX44QF,United Kingdom)
出处
《计算机应用》
CSCD
北大核心
2022年第S01期201-207,共7页
journal of Computer Applications
基金
国家自然科学基金资助项目(62072108)
福建省自然科学基金杰青项目(2020J06014)。
关键词
云计算
基于云的软件服务
资源分配
深度强化学习
反馈控制
cloud computing
cloud-based software services
resource allocation
Deep Reinforcement Learning(DRL)
feedback control
