面向云数据中心的虚拟机部署时延优化算法研究

发布时间：2018-07-02 09:25

本文选题：云计算 + 数据中心　；参考：《计算机应用研究》2017年12期

【摘要】：考虑了服务器内资源容量及虚拟机多类型资源需求时虚拟机部署最优化时延问题。首先将最优化虚拟机部署时延问题进行了形式化建模,并证明了该问题为一个NPC问题;然后通过遗传结合贪心策略提出了一种高效的虚拟机部署算法优化时延。该算法的主要特点在于:结合了贪心策略指导个体在初始化、选择、交叉、变异时形成最优解,并且在交叉过程中采用奇、偶数位个体交叉的方式形成新个体,既避免了个体间的重复交叉,又通过多样化的新个体形成使得算法的解不会陷入局部最优。另外,由于遗传算法在交叉过程中会存在交叉冲突问题(服务器容量超载),还设计了一种交叉冲突检查方法,避免了交叉冲突后非法个体的生成。最后,通过实验比对,将提出的算法分别与最新研究成果VMPDN、粒子群优化算法等进行比较,结果表明提出的算法有效地缩短了虚拟机的部署时延。同时通过不同资源类型数量、迭代次数及种群大小的情况下,分析和考察了算法性能,结果表明提出的算法性能仍优于其他的算法。
[Abstract]:The optimal delay problem of virtual machine deployment is considered when the resource capacity in the server and the multi-type resource requirements of the virtual machine are considered. Firstly, the problem of optimal virtual machine deployment delay is formalized and proved to be a NPC problem, and then an efficient virtual machine deployment algorithm is proposed by combining genetic strategy with greedy strategy. The main features of the algorithm are that the greedy strategy is used to guide the individual to form the optimal solution in initialization, selection, crossover and mutation, and in the process of crossover, the odd and even digit individuals cross to form a new individual. It not only avoids the overlapping between individuals, but also makes the solution of the algorithm not fall into local optimum through the formation of new individuals. In addition, because the genetic algorithm has the problem of crossover conflict (server capacity overload) in the process of crossover, a cross-conflict checking method is designed to avoid the generation of illegal individuals after cross-conflict. Finally, the proposed algorithm is compared with VMPDN and particle swarm optimization algorithms, respectively. The results show that the proposed algorithm can effectively shorten the deployment delay of virtual machine. At the same time, the performance of the proposed algorithm is analyzed and evaluated under the conditions of different resource types, iterations and population size. The results show that the proposed algorithm is still superior to other algorithms.
【作者单位】：桂林理工大学嵌入式技术与智能信息处理广西高校重点实验室;桂林理工大学信息科学与工程学院;
【基金】：广西自然科学基金资助项目(2015GXNSFBA139260) 国家自然科学基金资助项目(61563012,61540054) 广西高校重点实验室主任基金资助项目(2016-01-05) 桂林理工大学博士启动基金资助项目(002401003456)
【分类号】：TP18;TP302

【相似文献】