摘要
针对部署在云中的应用多而繁杂并且不同的应用对特定的资源呈现不同的敏感性问题,提出了一种基于主模式方法的云应用分类架构,能够比较精确地将应用分为CPU密集型、内存密集型、网络密集型和I/O密集型等类型,从而能够更好地对云中的资源进行调度;对于云中的应用对资源的消耗,提出了一种基于差分自回归移动平均(ARIMA)模型的预测算法,能够以低的预测误差(高预测平均误差7.59%,低预测平均误差6.06%)对消耗资源预测;对传统的基于虚拟化的应用云架构进行适当的修改,能够细粒度地应对应用的自动扩张,从理论上解决了基于虚拟机的资源提供的不灵活以及低效的问题。
Considering the applications deployed in the cloud which are rather complicated and different applications exhibit different sensitivity to issues of specific resources, an architecture based main mode method was proposed to classify applications into CPU-intensive, memory-intensive, network-intensive, and I/O-intensive precisely, enabling better scheduling of resources in the cloud; An ARIMA ( AutoRegressive Integrated Moving Average) model-based prediction algorithm, which was also implemented, can lower average prediction error (7.59% high average forecast error and 6.06% low average forecast error) when forecasting consumption of resources; Appropriate modifications have been made on the traditional virtualization-based application cloud architecture to solve the inflexibility and inefficiency of the architecture based on virtual machine.
出处
《计算机应用》
CSCD
北大核心
2013年第6期1534-1539,共6页
journal of Computer Applications
关键词
云计算
应用分类
资源预测
细粒度
资源提供
cloud computing
application classification
resource prediction
fine-grained
resource provision