Joint optimization of function mapping and preemptive scheduling for service chains in network function virtualization

https://doi.org/10.1016/j.future.2017.12.021Get rights and content

Highlights

  • We study the joint optimization on functions mapping and preemptive scheduling in NVF, with the objective to minimize the total completion time.

  • We formulate the optimization problem as an Integer Linear Programming (ILP).

  • An efficient online algorithm is also proposed aim at addressing the high computational complexity of solving ILP.

Abstract

The idea of Network Function Virtualization (NFV) is to decouple of network functions from dedicated hardwares to obtain higher flexibility in terms of network management and maintenance. Although the problem of function placement and scheduling in NFV has drawn much attention in recent years, existing studies only consider the precedence constraints between network functions within a service but lacking for the interactions among services. In this paper, under the premise of satisfying the precedence conditions of forming the service chain, we study how to minimize the completion time of the whole system through efficient mapping and preemptive scheduling of functions for multiple service chains. We firstly model this issue as an Integer Linear Programming (ILP) problem. To avoid the computational complexity of ILP, an online preemptive algorithm is designed. Extensive simulations are conducted to validate the effectiveness of the proposed algorithm. The simulation results indicate that our algorithm is suitable for placement and scheduling of functions for multiple service chains in NFV.

Introduction

In recent years, growing customer demands make Internet Service Providers (ISPs) into a dilemma [1]: on one hand, traffic volume keeps growing up sharply, burdening ISP with cost on capital expenditure (CAPEX) and operating expenditure (OPEX); on the other hand, the obtained profit per subscribed user tends to decrease due to the competition among ISPs. In such conditions, ISPs need to rebuild the current mobile network architecture to improve the network throughput, as well as to reduce both CAPEX and OPEX.

Network Function Virtualization (NFV) [[1], [2]] has been proposed as a novel paradigm aiming at tackling those challenges for ISPs. NFV decouples network functions/middleboxes [3] (e.g., load balancing, deep-packet inspection, firewalls, etc.) from dedicated hardware. In an NFV environment, network functions/middleboxes are abstracted into software modules, called Virtual Network Functions (VNFs), running on commercial-off-the-shelf (COTS) hardware such as commercial servers and switches [4]. By stripping network functions from specific hardware, NFV not only reduces CAPEX and OPEX [5], but also significantly improves the flexibility and fault tolerance. In addition, by sharing the hardware among VNFs, it significantly improves the utilization of network resources. The advantages of NFV has attracted much attention and can be widely applied to fields like big data [[6], [7]], cloud computing [8], mobile computing [9].

Usually, network functions do not exist isolatedly, but are chained together to provide a specific service, namely service chain. The functions in a service chain are executed orderly with a predefined sequence. Although NFV provides huge potential benefits for ISP, there are still challenges that need to be addressed. The implementation of decoupling network functions from the dedicated hardware is the first challenge to apply NFV as the network infrastructure. The requirement and architecture standards are proposed by several industry groups, including the NFV industry standards group (ISG) of the European Telecommunications Standards Institute (ETSI) [[10], [11]] and the Internet Engineering Task Force (IETF) [12]. Another challenge exists in the issue of how to map and schedule functions of a service chain to hardware devices (e.g., servers). For this issue, the most common practice is to deploy each VNF in a dedicated virtual machine (VM). For example, Zhang et al. [13] propose an online stochastic auction mechanism for on-demand service chain provisioning and pricing for an NFV provider. It considers each VM is referred to as an instance of a VNF. In addition, we notice that existing studies only consider the precedence constraints among network functions in a single service chain, but leave the relationship among different service chains out of consideration.

Therefore, we are motivated to study the joint optimization of VNF mapping and scheduling for multiple service chains in NFV environment. In such scenario, a VM can handle more than one network functions. A VM caches all the requested functions in a queue and executes them sequentially. When another function is requested by a new service chain, we schedule the newly requested one to a suitable position of the queue, not simply adding it at the end of the queue. As a result, the ongoing service chain can be interrupted in order to reduce the completion time for all the service chains. We formulate this problem as an ILP with the objective to minimize the total completion time for the given service chains. In order to reduce the computational complexity for solving the problem, we propose an online algorithm which dynamically adjusts the execution order of VNFs deployed in one VM. To our best knowledge, we are the first to study the VNF mapping and scheduling for NFV, with the considerations of preemptive scheduling for multiple VNFs on one VM. Our contributions are as the following two folds:

  • We study the joint optimization on functions mapping and preemptive scheduling in NVF, with the objective to minimize the total completion time.

  • We formulate the optimization problem as an Integer Linear Programming (ILP). An efficient online algorithm is also proposed to address the computational complexity of solving ILP. Experiment results validate the correctness and efficiency of our proposal.

The rest of the paper is organized as follows: Section 2 summarizes the state-of-art work on NFV. Section 3 gives the system model, and Section 4 formulates the completion time minimization problem into an ILP. An efficient online algorithm for the ILP is proposed in Section 5. The performance evaluation results are given in Section 6. The paper is concluded in Section 7.

Section snippets

Related work

The NFV technology has drawn much attention from both academia and industry in recent years, including fields of cloud computing  [[14], [15]], edge computing [16], parallel computing [17], cyber–physical system [18], sensor network [19]. In cloud computing, a chain of operations are usually required. For example, Shen et al. [20] propose a framework to secure the urban data sharing, which requires the separation of data flow and security parameter flow. Fu et al. [21] design a conceptual graph

System model

In this paper, we consider a network where network entities are implemented by virtual node equipped with computational resource forming the virtual node set N. Any virtual node jN can host one or several network functions, while it can execute at most one instance for a given time. Let F indicate the set for all virtual functions in the system, and binary variable βi,j denote whether function iF can be handled by node j (βi,j=1) or not (βi,j=0).

A virtual node in the network has a capacity

Problem formulation

In this section, we formulate the minimization of the total completion time for all service chains as an ILP problem. All parameters and variables used in the formulation and analysis have been listed in Table 2.

Algorithm design

There are a lot of various optimizer, e.g., Gurobi Optimizer,1 that can obtain the optimal solution for the ILP problem, albeit with high computational complexity. Aiming at reducing the computational complexity, we design an efficient preemptive algorithm based on greedy strategy. Whenever a service chain arrives, we execute the algorithm to schedule the functions in the requested service chains based on the current state, as shown in Algorithm 1.

For each function in a

Evaluation and analysis

In order to evaluate the proposed preemptive algorithm for mapping and scheduling of VNFs in multiple service chains, we conduct extensive simulation based experiments, by varying node buffer size, the number of nodes, function buffer requirement, function coverage on nodes and the number of functionsin service chains. The default settings of our simulations are listed in Table 3. Our proposed algorithm is compared with three competitors, i.e., the optimal solution by Gurobi Optimizer, Shortest

Conclusion

Function placement and scheduling has already become a widely concerned issue in NFV. Not only precedence constraints between network functions in a service chain, but also the interaction between service chains should be taken into consideration. In this paper, we study how to minimize the completion time of the whole system through efficient mapping and preemptive scheduling of functions for multiple service chains. We model the mapping and scheduling problem in the form of ILP and propose an

Acknowledgments

This research was supported by the NSF of China (Grant No. 61673354, 61672474, 61602199, 61772480, 61402425), the Provincial Natural Science Foundation of Hubei (Grant No. 2016CFB107, 2015CFB400 ). This paper has been subjected to Hubei Key Laboratory of Intelligent Geo-Information Processing, China University of Geosciences, Wuhan 430074, China.

Hong Yao received the B.E. degree in Computer and Applications from Wuhan Technical University of Surveying and Mapping in 1998, the M.E. degree in Cartography and Geographic Information Engineering from China University of Geosciences, Wuhan, and the Ph.D. degree in Computer Science and Technology from Huazhong University of Science and Technology in 2010. He is currently an associate professor in the School of Computer Science at China University of Geosciences, Wuhan. His research interests

References (36)

  • WangL. et al.

    Link the remote sensing big data to the image features via wavelet transformation

    Cluster Comput.

    (2016)
  • HeZ. et al.

    Decomposition tree: a spatio-temporal indexing method for movement big data

    Cluster Comput.

    (2015)
  • KhoshkbarforoushhaA. et al.

    Dimensions for evaluating cloud resource orchestration frameworks

    Computer

    (2016)
  • YaoH. et al.

    Opportunistic offloading of deadline-constrained bulk cellular traffic in vehicular DTNs

    IEEE Trans. Comput.

    (2015)
  • N. ETSI, ETSI network functions virtualisation (NFV) industry standards (ISG) group draft specifications, 2014,...
  • HawiloH. et al.

    NFV: state of the art, challenges, and implementation in next generation mobile networks (vEPC)

    IEEE Netw.

    (2014)
  • J. Halpern, C. Pignataro, Service function chaining (sfc) architecture, Tech. Rep.,...
  • ZhangX. et al.

    An online stochastic buy-sell mechanism for VNF chains in the NFV market

    IEEE J. Sel. Areas Commun.

    (2017)
  • Cited by (0)

    Hong Yao received the B.E. degree in Computer and Applications from Wuhan Technical University of Surveying and Mapping in 1998, the M.E. degree in Cartography and Geographic Information Engineering from China University of Geosciences, Wuhan, and the Ph.D. degree in Computer Science and Technology from Huazhong University of Science and Technology in 2010. He is currently an associate professor in the School of Computer Science at China University of Geosciences, Wuhan. His research interests include wireless and mobile network, delay tolerant networks, overlay networks, and mobile cloud computing.

    Muzhou Xiong received his Ph.D. and B.E. degree both n computer science from Huazhong University of Science and Technology, Wuhan, China, in 2010, 2002. He is currently an associate professor with School of Computer Science of China University of Geosciences, Wuhan, China. His research interests include wireless and mobile network, high performance computing, and complex system modeling and simulation.

    Hui Li is currently a master student at school of computer science with China University of Geosciences. His research areas include software-defined sensor networks and network function virtualization.

    Lin Gu received her M.S. and Ph.D. degrees in computer science from University of Aizu, Fukushima, Japan in 2011 and 2015. She is currently a lecturer in School of Computer Science and Technology, Huazhong University of Science and Technology, China. She is a member of IEEE. Her current research interests include cloud computing, vehicular cloud computing, Big Data and Software-defined Networking.

    Deze Zeng received his Ph.D. and M.S. degrees in computer science from University of Aizu, Aizu-Wakamatsu, Japan, in 2013 and 2009, respectively. He received his B.S. degree from School of Computer Science and Technology, Huazhong University of Science and Technology, China in 2007. He is currently an associate professor in School of Computer Science, China University of Geosciences (Wuhan), China. His current research interests include: cloud computing, software-defined sensor networks, data center networking, networking protocol design and analysis.

    View full text