Recently, network services are increasingly connecting computational elements within and across datacenters. In the Network Functions Virtualization (NFV) environment, to successfully orchestrate a network service, first a VNF-Forwarding Graph (VNF-FG) must be composed that realizes the required functionality. Second, this VNF-FG must be embedded onto the infrastructure, that is increasingly becoming heterogeneous. To efficiently allocate service demands, intelligent mechanisms and algorithms are needed to effectively tailor the VNF-FG to the cloud network onto which the service will be deployed. This paper introduces the first service model supporting on-demand orchestration of services with bidirectional chaining and delay constraints, optional VNFs and traffic aggregation. Building on this service model, we propose algorithms looking for the optimization of the order and number of VNF instances, to adapt the VNF-FG to the available resources in the substrate network. Numerical experiments show that, through coordination of composition and embedding tasks, our proposed algorithms can significantly improve the acceptance ratio, compared to algorithms that perform these tasks in two separate stages.

最近，网络服务越来越多地连接数据中心内和数据中心之间的计算元素。在网络功能虚拟化（NFV）环境中，要成功协调网络服务，首先必须组成一个实现所需功能的VNF转发图（VNF-FG）。其次，该VNF-FG必须嵌入到越来越多样化的基础架构中。为了有效分配服务需求，需要智能机制和算法来有效地将VNF-FG调整为将要在其上部署服务的云网络。本文介绍了第一个服务模型，该模型支持具有双向链接和延迟约束，可选VNF和流量聚合的按需业务编排。基于此服务模型，我们提出了寻找VNF实例的顺序和数量优化的算法，以使VNF-FG适应底物网络中的可用资源。数值实验表明，与在两个单独阶段执行这些任务的算法相比，通过协调构图和嵌入任务，我们提出的算法可以显着提高接受率。