LBBSRT: An efficient SDN load balancing scheme based on server response time

https://doi.org/10.1016/j.future.2016.10.001Get rights and content

Abstract

The response time is the most important factor determining user experiences in the service provision model involving server clusters. However, traditional server cluster load balancing scheme are limited by the hardware conditions, and cannot completely exploit the server response times for load balancing. In order to effectively resolve the traditional load balancing schemes, we propose a load balancing scheme based on server response times by using the advantage of SDN flexibility, named LBBSRT. Using the real-time response time of each server measured by the controller for load balancing, we process user requests by obtaining an evenly balanced server loads. Simulation experiments show that our scheme exhibits a better load balancing effect and process requests with a minimum average server response times. In addition, our scheme is easy to implement, and exhibits good scalability and low cost characteristics.

Introduction

Achieving optimum load balancing is of significant importance whilst combating network overhead issues in any distributed processing architectures. Service availability is paramount in measuring end user satisfaction  [1], which is heavily impacted by the level of achievable load balancing among the process clusters. In general, a well-balanced load in the network helps to optimize the utilization of the available resource by the ways of maximizing the throughput, minimizing the response time, and avoiding overloading resources in the network  [2]. For the purposes of alleviating heavy-traffic network flux and reducing the risk of single server becoming the main overhead contributor, many datacentres adopt dedicated hardware resources to achieve load balancing whilst supporting a large number of users  [3]. However, the increasing costs and technical complications in the deployment of such hardware systems often require human intervention to ensure consistent functioning of such strategies  [4].

Software-Defined networking (SDN) is one of the notable forms of computer networking  [5], [6], facilitating a simple and conveniently maneuverable network flow control method requiring minimal investment costs whilst availing maximum benefits for a massive number of users. SDN controls the data transportation by deploying the network switches as a software implementation, whereby a flow table lookup operation will be carried out whenever a data flow arrives at the switches. Flow tables  [7] ([Header: Counters: Actions]) are widely used in SDN. The headers and counters of the flow table are updated accordingly whenever actions relevant to flow changes are imposed. During this update process, the header information is usually recorded onto the database and the OpenFlow switches process the data flow in accordance with the header records. Based on the SDN model with a centralized controller, an OpenFlow switch  [8] is designed with different rules to control the network traffic using the header records. Balancing the network load at the software tier is now practically realizable using the SDN facilitated flow control system. To this end, Handigol  [9] proposed plug, a load balancing model based on SDN. Based on the Openflow environment, Kaur  [10] achieved network load balancing using polling algorithm. Further, Zhang  [11] achieved the minimum number of connections in the network using the polling algorithm of load balancing under the SDN framework. Shang  [12] incorporated a middlebox based on the SDN architecture to achieve load balancing by collecting the server information. Despite the existing implementations of SDN to resolve high cost and poor flexibility issues in achieving effective load balancing, notable drawbacks are still prevalent in the aforementioned schemes. To add a few, Kaur and Zhang  [10], [11] applied traditional load balancing algorithms to the SDN architecture, and so the two schemes cannot effectively reduce the server response time. Though Shang  [12] can effectively reduce the server response time, this scheme relies on the server information which increases the complexities of the server architecture.

This paper proposes a new method of load balancing in SDN networks with the motivation of enhancing the load balancing effect by reducing the server response time. In this paper, the server response time is defined by the interval that begins from accepting user requests to responding to user requests for server. If servers in a server cluster have several similar performances and provide the same service, then for each server, the higher the load is, the longer the response time is. Correspondingly, the longer the response time is, the higher the corresponding load is. Therefore we propose a load balancing scheme based upon server response time. It can solve the problem of the load balancing in the server cluster based on the server response time. Our proposed approach effectively overcomes the drawbacks of the traditional methods, including high cost, low reliability and poor extensibility. The contributions of the paper include:

  • An effective load balancing scheme based on SDN architecture, using the real-time response time of each server measured by an SDN controller.

  • Realizing the potential implementation of our scheme by incorporating a floodlight controller module in the scheme.

  • Proving the effectiveness of our proposed scheme by evaluating the response time and resource utilization metrics against the traditional schemes.

The rest of the paper is organized as follows: Section  2 reviews the existing traditional load balancing schemes and introduces the background of SDN. Section  3 details the design of our proposed scheme, LBBSRT (Load Balancing by Server Response Time). The performance evaluation of LBBSRT is presented in Section  4. Section  5 concludes the paper.

Section snippets

The traditional load balancing scheme

The traditional load balancing schemes are categorized into four major types  [12] such as based on the client, based on the middle layer, based on the DNS, and based on the transport layer.

In the load balancing scheme based on the client side, clients primarily collect every server running parameters from the server clusters either periodically or non-periodically, and send a request to different servers to achieve load balancing. Although this method can achieve a certain degree of load

The design and implementation of LBBSRT

Usually, obtaining the response times of each server from a pool of servers is a tedious process using the traditional network equipment. Such traditional schemes do not incorporate the server response times whilst balancing the server loads. Instead, they simply ping the servers for obtaining their reply time. Such strategies may not obtain the actual reply time of the server and so are often not accurate. The segregation of the control plane and data plane in the SDN facilitates obtaining the

Experiment result and performance analysis

In our experimental setup, the virtual switch is created by Open vSwitch. The floodlight is chosen as the SDN controller. Due to the floodlight is a free open source, and add and delete modules can be arbitrary used, so it provides much more convenience for our test. Three virtual machines with identical configurations are assigned as servers to provide web services. In this experiment, we let 30 clients to access to the server. Moreover, the access frequencies of different clients are usually

Conclusions and future work

The emergence of SDN architecture provides us with a new train of novel prospects for solving the prevailing issues in the traditional load balancing network. In order to solve the problems of lower efficiency and higher deployments costs of load balancing in the traditional networks, this paper proposes a dynamic load balancing scheme under the SDN architecture, using the controller to obtain the real response time of each server ultimately to select a server with minimum or the most stable

Acknowledgments

The work was supported by the National Natural Science Foundation of China (No. 61572001, No. 61502008), the Research Fund for the Doctoral Program of Higher Education (No. 20133401110004), the Educational Commission of Anhui Province, China (No. KJ2013A017), the Natural Science Foundation of Anhui Province (No. 1508085QF132), the Tender Project of the Co-Innovation Center for Information Supply & Assurance Technology of Anhui University (No. ADXXBZ2014-7), and the Doctoral Research Start-up

Hong Zhong is a Professor (from 2009) and Executive Dean of the School of Computer Science and Technology, Anhui University, China. She received Ph.D. degree in University of Science and Technology of China in 2005. Her research interests cover network and information security.

References (27)

  • S. Huang et al.

    Network Hypervisors: Enhancing SDN Infrastructure

    Comput. Commun.

    (2014)
  • M. Kobayashi et al.

    Maturing of OpenFlow and Software-defined Networking through deployments

    Comput. Netw.

    (2014)
  • F. Pakzad et al.

    Efficient topology discovery in OpenFlow-based Software Defined Networks

    Comput. Commun.

    (2016)
  • Z. Huang et al.

    A threshold-based multi-traffic load balance mechanism in LTE-A networks

  • J. Zha et al.

    Research on load balance of service capability interaction management

  • W. Tian et al.

    Prepartition: A new paradigm for the load balance of virtual machine reservations in data centers

  • E. Musoll

    Hardware-based load balancing for massive multicore architectures implementing power gating

    IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst.

    (2010)
  • F.A. Lopes et al.

    A software engineering perspective on SDN programmability

    IEEE Commun. Surv. Tutor.

    (2016)
  • F. Hu et al.

    A survey on software-defined network and openFlow: from concept to implementation

    IEEE Commun. Surv. Tutor.

    (2014)
  • P. Bosshart et al.

    Forwarding metamorphosis: Fast programmable match-action processing in hardware for SDN

    ACM SIGCOMM Comput. Commun. Rev.

    (2013)
  • J. Naous et al.

    Implementing an OpenFlow switch on the NetFPGA platform

  • N. Handigol et al.

    Plug-n-Serve: Load-balancing web traffic using OpenFlow

    ACM Sigcomm Demo

    (2009)
  • S. Kaur et al.

    Round-robin based load balancing in Software Defined Networking

  • Cited by (0)

    Hong Zhong is a Professor (from 2009) and Executive Dean of the School of Computer Science and Technology, Anhui University, China. She received Ph.D. degree in University of Science and Technology of China in 2005. Her research interests cover network and information security.

    Yaming Fang is now a research student in the School of Computer Science and Technology, Anhui University. His research interest is Software Defined Networking.

    Jie Cui is now an Associate Professor in the School of Computer Science and Technology, Anhui University. He received Ph.D. degree in University of Science and Technology of China in 2012. He has published over 30 papers. His research interests include network and information security.

    View full text