A Study on Efficient Service Function Chain Placement in Network Function Virtualization Environment
Yansen Xu · Institutional Repositories DataBase (IRDB)
Traditional network is equipped with many proprietary hardware-based network functions.These hardware-based network functions lead to network ossification which results in lack of flexibility and high cost to network operators.Network function virtualization (NFV) is a promising technology to solve this ossification problem in traditional network.NFV separates software implementation of network function from hardware.The software implementation of network function is called virtual network function, which can be installed and run on generic servers.A service function chain (SFC) in NFV is a set of ordered virtual network functions (VNFs) for end-to-end network service.The core technology to realize network service in NFV is SFC placement.SFC placement deals with placing VNFs of an SFC onto which hardware server, and chaining these VNFs along which edges in the hardware platform while satisfying particular requirements of SFC.Due to limited network resources and requirements of network service performance, network operators must carefully design scheme of SFC placement to efficiently utilize network resources and guarantee network service performances.In this thesis, we focus on SFC placement problem with satisfying minimized cost, optimal end-to-end delay, and distributed placement requirements.These requirements are critical to network operators constructing network service, and end users using network services.We contribute a mathematic model and algorithms to solve these problems.From the aspect of cost, considering cost is one of critical aspects to network operator, we formulate a mathematic model with objective of minimizing setup costs and operational costs.With a realistic definition of setup cost and operation cost, our model can reduce cost up to 30% in average.Moreover, our model also considers guaranteeing the end-to-end delay deadline, which is a critical requirement to SFCs, such as real-time applications.Our model can satisfy 100% end-to-end delay deadline requirements of SFC requests.