摘要
云计算环境为用户提供了大量弹性可扩展的基础设施资源,用户可以按需购买和支付服务.如果云平台能够自动分配给一批任务合理规模的计算资源,将会方便用户使用并且较大节省用户服务成本.本文在瀚海星云云计算平台上构建了一个软件应用服务层,提供Bag-of-tasks(BoT)应用公共服务.根据历史信息通过回归方法和BP神经网络方法对BoT中的子任务进行执行时间预测.然后使用最大并发度概念,在虚拟机内存是否满足任务情况下,提出了VMA,NP-IO和NP-DP三种算法.最后,使用图片分割软件作为BoT应用,从资源分配情况、任务完成率和算法时间复杂度方面验证了算法的有效性.
Cloud computing service provides users with unlimited and elastic infrastructure resources, and users can buy and pay the service as they demand. If Cloud platforms can automatically assign a reasonable scale of computing resources to tasks, it will save service cost for users. In this work, a software service platform that can support Bag-of-tasks （ BoT ） application is built on the Cloud infrastructure service platform. The execution time of subtasks in BoT will be estimated by the combined method of regression analysis and BP neural network after the study of historical information. Meanwhile, we use the idea of the degree of concurrency to maximize the performance of VMs. And then, VMA, NP-IO and NP-DP algorithms are proposed in the resource allocation strategy under the fact that different tasks need various memory capacities. Finally, we use image segmentation software as the BoT application and evaluate the algorithm efficiency in three ways of resource utility, task completion rate and time-complexity.
出处
《小型微型计算机系统》
CSCD
北大核心
2014年第1期12-18,共7页
Journal of Chinese Computer Systems
基金
广东省中国科学院全面战略合作项目(2010A090100027)资助
