数据中心应用感知的动态资源配置研究

Application-aware dynamic resource allocation in data center

面向云计算数据中心,该文研究了应用感知的动态资源配置及其实现方法。采用马尔科夫过程对应用进行建模,并计算各功能模式下资源需求的变化关系。以保证应用执行性能为前提,提出基于执行等待时间期望的资源评估算法。提出应用感知的动态资源配置算法并和静态最大资源配置算法进行了对比实验。实验结果表明,在相同的应用负载下,该文算法相对于最大资源配置算法能够减少49%的物理机使用量;在给定同样的物理机数时,该文算法对应用请求的接受率比最大资源配置算法提高约45%。

A new dynamic resource allocation algorithm is researched for data center oriented to cloud computing. Applications are modeled using Markov process method,and the change pattern of resource requirements inside the applications is calculated. A resource evaluation approach is proposed based on the expectation of execution waiting time to guarantee application execution performance. An application-aware dynamic resource allocation algorithm is proposed and compared with static max resource allocation algorithm. The experimental results show that,this algorithm reduces the resource consumption by 49% under a same application load,improves the acceptance rate of application request by about 45% for a same given physical server number.