Network Virtualization enables service providers to instantiate virtual networks while sharing the same physical infrastructure. Virtual networks are allocated to clients for deploying their services. In this context, clients are served without knowing which server is replying and which routing strategy is adopted. In this paper, we focus on the server selection problem for competitive service providers in Network Virtualization context. These latter do not own the physical infrastructure and aim to minimize the cost of leased resources while meeting their clients’ SLA (Service Level Agreement) targets. We propose a mixed integer linear program whose objective is to minimize latency under clients’ demand and server’s bandwidth constraints based on low and medium traffic intensities. We show that concentrating the traffic on the closest server yields an optimal solution. In addition, through simulation and analytical results, we show that Traffic Concentration reduces the response time, jitter, and node and link utilization compared to YouTube, random server selection, and equal load distribution of traffic among servers. However, the latter technique yields the best performance in terms of packet loss.

网络虚拟化使服务提供商可以实例化虚拟网络，同时共享相同的物理基础架构。将虚拟网络分配给客户端以部署其服务。在这种情况下，在不知道哪个服务器正在答复以及采用哪种路由策略的情况下为客户端提供服务。在本文中，我们专注于网络虚拟化环境下竞争服务提供商的服务器选择问题。后者不拥有物理基础架构，其目的是在满足客户的SLA（服务水平协议）目标的同时，将租赁资源的成本降至最低。我们提出了一个混合整数线性程序，其目的是基于中低流量强度，在客户端需求和服务器带宽约束下最大程度地减少延迟。我们表明，将流量集中在最接近的服务器上可产生最佳解决方案。此外，通过仿真和分析结果，我们发现与YouTube相比，“流量集中”减少了响应时间，抖动以及节点和链接的使用率，减少了服务器的随机选择，并在服务器之间平均分配了流量。但是，后一种技术在数据包丢失方面产生了最佳性能。