重力模型的数据中心网络流量推理算法

Traffic Inference for Data Center Network Base on Gravity Model

数据中心网络是云计算等大型分布式计算服务的基础,有效地设计与管理数据中心网络需要遵循该网络的流量特征。而目前直接对数据中心网络进行端到端地流量测量是非常困难的,间接地通过SNMP数据推理得到端到端流量的方法已在传统计算机网络中得到认可,但无法直接应用于现有的数据中心网络。为了解决以上问题,提出一种基于重力模型的数据中心网络流量推理算法,首先根据数据中心网络流量的条件独立性将网络拓扑分解为若干子集,在此基础上提出相关定理可准确地计算出网络中的粗粒度流量,最后利用重力模型和网络层析技术得到细粒度端到端流量。通过与现有的流量推理算法SRMF和ELIA在NS3搭建的不同规模的数据中心网络中做性能对比,实验结果表明新算法能有效地利用数据中心拓扑结构特点,在保证计算效率的前提下,将计算准确度大幅提升,可满足当前数据中心网络实时获取端到端流量数据的需求,为今后数据中心网络的设计和研究提供了重要参考依据。

Data Center Network （DCN） is the infrastructure of cloud computing and other distributed computing services. Understanding the chardcteristics of end-to-end traffic flows in DCNs is essential to DCN designs and operations. However,it is extremely difficult to measure the traffic flows directly. Inferring the end-to-end traffic follows from the SNMP counters on switches has been widely applied in traditional computer networks. But it still can not be utilized in DCNs directly yet. To address this problem,we propose an efficient traffic inference algorithm for DCNs based on gravity traffic model. It first decomposes the DCNs into several clusters according to the feature of conditional independence of DCN traffic,and then computers the coarse-grained traffic of each cluster based on some theorem that we state in section 3. Finally it utilizes the gravity traffic model and network tomography to refine the traffic on each cluster to obtain the fine-grained end-to-end traffic. We compare our new proposal with two classical traffic inference algorithms SRMF and ELIA on different scale of DCNs. The results show that our new algorithm outperforms the other two algorithms in both speed and accuracy. Thus the new proposal can provide vital reference to the area of DCN designs and operations.