一种基于短期预测的业务弹性伸缩算法

摘要

Aiming at the service deployment problem of geographically distributed clouds,a Short-term Prediction based Elastic Service Scaling ( SPESS ) is proposed. The algorithm predicts the arrival rate of user requests by Dynamic Autoregressive Integrated Moving Average( D-ARIMA) model. In comprehensive consideration of the prediction results, current load and processing rate of the service,the number of virtual machines in every site is decided to strike a balance between quality of service and cost. Experimental results show that the algorithm can ensure better balance between quality of service and cost,and the load of service,transmission delay and queueing time are controlled in a reasonable range.%针对地理分布云的业务部署问题，提出一种基于短期预测的业务弹性伸缩算法SPESS。该算法利用动态差分自回归移动平均模型对用户请求到达速度进行预测，综合考虑预测结果、业务的当前负载及处理速度，调整每个站点虚拟机的数量，从而在保障服务质量的同时尽可能地降低运行成本。实验结果表明，该算法能够在保障服务质量和运行成本之间取得较好的平衡，且业务整体负载、传输延时、排队时间均控制在一个合理的范围内。