Jointly Optimized Placement of Application VM and VNF in NFV Based Data Center

In data centers, applications are typically deployed in a distributed manner across servers via Virtual Machines (VMs). In order to enhance the application security and performance, the data flows between VMs often traverse a particular set of middleboxes (such as Firewall, NAT, etc.) in a predefined sequence. One of the key challenges that data centers have been facing is how to efficiently place VMs to ensure the application performance and improve the resource efficiency of data centers. On the other hand, Network Function Virtualization (NFV) decouples middlebox function (also called network function) software from specified appliances, and deploys it onto general shared servers by virtualization technology. It has been being regarded as a promising technology to overcome high Capital Expenditures (CAPEX) and Operational Expenditures (OPEX) on the middlebox deployment and maintenance. In NFV, the VM deployed with network function software is called VNF (Virtualized Network Function). When NFV technology is applied in data centers, the locations of VNFs affect the resource efficiency of the data centers as well. To distinguish the VMs hosting the VNFs, the VMs hosting the application business are called Application VMs (AppVMs). As the endpoints of the data flows, the locations of the AppVMs determine the successful deployment of the required VNFs between the AppVMs to a large extent. Therefore, in NFV enabled data centers, it is necessary to study the joint optimization of the AppVM and VNF placement. However, almost all existing studies have ignored it. This paper is the first to deal with the problem of the joint optimization problem of AppVM and VNF placement. Firstly, we model the problem of the joint optimization problem of AppVM and VNF placement as a binary integer linear programming model. Next, due to the NP hard characteristic, we propose two joint optimization methods of AppVM and VNF placement. Finally, through a large number of experiments, compared with the algorithms that deal with the AppVM and VNF placement separately, the advantages of the proposed joint optimization methods in improving the request acceptance rate are verified.