气象集约化资源池计算资源容量估计方法研究

Research on Computing Resource Scale Estimating Method for Intensive Meteorological Resource Pool

精确估计计算资源的规模是气象集约化资源池合理设计的关键,但是由于应用对计算资源需求可能突发增长,应用部署方案种类繁多等因素的存在,致使计算资源规模估计困难。为解决该问题,本文对计算资源容量规划问题进行建模,并在分析问题求解难度的基础上,根据问题的特点设计了一种基于离散差分进化算法的计算容量估计方法。在问题建模上,本文采用资源预留策略来保证应用能获取足够的计算资源。在方法设计上,方法首先建立了应用的部署顺序与计算资源容量规模的映射关系,并以所需参考服务器的规模和资源利用水平来综合评估应用部署顺序的适应度;然后,通过随机方式生成初始方案,并以适应度为评价标准,利用变异、交叉、局部搜索等操作改变应用的部署顺序;最后,通过迭代搜索估计计算资源所需的规模。实验结果表明该方法能有效估计计算资源容量的规模。

Estimating the scale of computing resource accurately is crucial for designing the intensive meteorological resource pool. However,it is difficult to estimate the scale,due to the growth of emergency demand for computing resource and a large number of schemes for deploying the meteorological application in different servers. In this paper,a model was built to describe the problem of estimating the scale of computing resource,and a differential evolution algorithm for estimating the computing resource scale was proposed to solve it by analyzing the difficulty of problem solving. To ensure the application of sufficient resources,reserving computing resource strategy was used in this model. To solve the problem,this algorithm firstly built a mapping relationship between application deployment order and computing resource scale,and a fitness function was proposed to evaluate the application deployment order by taking account of the number of servers and the resource utilization. Then,the initial application deployment orders were generated randomly,and they changed by mutation operation,crossover operation and local search operation according to their fitness. At last,the scale of computing resource was estimated after the iterative search. The experiment result shows that this algorithm can estimate the computing resource scale effectively.