A survey on energy efficiency in software defined networks

Abstract

Wide deployment and dense usage of computer networks may cause excessive energy consumption due to the increase in probability of network congestion, frame collisions and packet dropping rates resulting from late-received frames. Besides, based on the dramatic increase in network complexity with wireless, mobile and split tunnel connections, weak visibility into network flows and high cost of some of public and private network services, current networks can also be implied as inefficient in terms of both performance and economy. Software-Defined Networking (SDN) is a novel networking architecture, which provides a directly programmable and (logically) centralized network control, separates network control from forwarding, and enables programmable network components. SDN can have a significant role in reducing the aforementioned excessive energy consumption caused by data centers, network components, and end hosts. In this paper,¹ we examine the principles, benefits, and drawbacks of up-to-date SDN approaches that focus on energy efficiency. We also provide a brief comparison of possible energy gain ratios of existing approaches, discussion on open issues and a guideline for future research.

Introduction

In current communication networks, setting up a network requires multiple software and hardware-based networking technologies, such as protocols, switches and routers. Therefore, wired and wireless communication networks, such as personal, local or wide area networks, may face with a fundamental complexity problem in case they consist of a high number of connected devices having different characteristics and requirements. Additionally, devices in traditional data networks need to be configured individually, task updates of devices are time consuming, and the evolution of the hardware functionality is controlled only by the provider. Therefore, implementing a new network policy may require configuring many devices. Furthermore, traditional data networks have devices that are designed to perform specific tasks, such as switches and routers. Hence, they have a static architecture and result in slow-evolving network functionality. However, there is a high amount of increase in the use of mobile devices, server virtualization, and cloud services. All of these activities require dynamic network structure.

In the past decade, network traffic pattern has also changed. Applications now require accessing different servers and databases since users want to reach applications, infrastructure and all other IT resources. Mobile devices are currently used to access the corporate networks; hence, IT staff becomes obliged to protect their data. In addition, data centers are growing. Although intercommunication between these centers has been evolving, the ever-growing communication demand is mostly addressed by increasing the bandwidth and setting up new network cables. As a result, handling with massive datasets is costly with traditional networks. Since network virtualization is now widely used, physical elements can contain more than one virtual network structure. However, the reconfiguration process is not agile, easy, and quick. Besides, users’ network functionality is limited; it depends on the vendors and the hardware providers. Consequently, all of the above-mentioned facts constitute novel challenges for network owners.

As the network size increases, its complexity grows exponentially. In this regard, current network infrastructure requires more flexible and dynamic network operations, programmability and easily modified network devices. Therefore, the network must be programmed according to its changing requirements. Currently, the key paradigm to achieve this behavior is Software-Defined Networking (SDN). SDN simply provides a (logically) centralized control point in the network. In an SDN architecture, control plane - the plane that understands the network and decides the flow paths, and data plane - the plane responsible for the transmission of packets, are separated. Separation of control layer from the data layer enables programmability, increases functionality, and provides remote management between infrastructures using a single open protocol. This structure allows network and business applications to work together with the help of analytics and to reconfigure the network policies according to the changing user experience and application performance. In this context, network design and architecture remain the same while applications and systems progress to an advanced level.

In SDN, network intelligence and state are logically centralized, and the underlying network infrastructure is abstracted from the applications [1]. Network devices merely need to accept instructions from the SDN controller instead of understanding all protocols. Hence, any network element can be changed instantaneously. SDN enables the network to meet instantly emerging needs of business and institutions and help them to customize the network. SDN can also be beneficial in cases like directing devices to safe flows, improving network performance, preparing network bandwidth for scheduled data transfers, keeping network slices apart with controller to keep away from research traffic, transition between multiple data centers and single data center, etc.

In SDN, communication between infrastructure and control layer is provided by the OpenFlow protocol [2], which was released by Stanford University and California University in 2011 and currently managed by the Open Networking Foundation (ONF). Although the terms SDN and OpenFlow are closely related, they are not interchangeable. While SDN is an emerging network architecture that enables networks to be programmable with a high degree of automation, OpenFlow is a protocol that configures network switches in order to provide the communication between the SDN controller and the devices. OpenFlow allows SDN controllers to decide the path of network packets through the network of switches. However, controllers should also include orchestration tools to dynamically and automatically respond to the needs of the network. Moreover, controllers should be able to communicate with other controllers regardless of their own proprietary interfaces and scripting languages.

In a nutshell, SDN technology aims to address the problems of the traditional networks and it is currently supported by the network community. It can provide centralized control of the network even when the network consists of multi-vendor elements. It reduces the complexity through network automation. Moreover, via SDN, IT staff can easily adapt the network to the rapidly changing needs and requirements. In addition, SDN enables more security since it provides a global control point over the network. SDN controller can easily regulate any policy; therefore, IT staff can control even the smallest elements of the network, such as devices, users, and applications.

Another inefficiency of the current networking technology is the high amount of energy it consumes. Current networks are inefficient both environmentally and economically (i.e. CO₂ emission, operational costs, etc.) and hence they should be reconfigured. In the past, research scope of the Information and Communication Technology (ICT) was mainly based on performance and cost. The research community put insufficient effort to the energy consumed by ICTs and their impact on the environment. Current trends, such as increasing electricity costs, reserve limitations, and increasing emissions of carbon dioxide (CO₂) are shifting the focus of ICT towards energy-efficient and well-performed solutions. Even though governments and companies are now aware of the massive carbon emissions and energy requirements, it is obvious that carbon emissions and the amount of energy consumption will continue to increase [3]. As stated by the SMART 2020 study [4], ICT-based CO₂ emissions are rising at a rate of 6% per year. With such a growth ratio, it is expected that CO₂ emissions caused by ICTs will reach 12% of worldwide emissions by 2020. Communication networks designed according to this energy efficiency criteria are called green networks [5]. In this context, SDN procedure can have a significant role in reducing the energy consumption by decoupling network functionalities and utilizing some of the local and network-related parameters. In other words, the global network knowledge and centralized decision-making mechanism of SDN make it a proper environment for realizing green networks.

There have been many reviews focusing on SDN paradigm in the literature, such as the works presented in [6], [7], [8], [9]. However, extensive analysis of SDN in terms of energy efficiency has hitherto received little attention. Towards closing this gap, we first present the importance and possible ways of energy saving in SDN, and then examine the principles, benefits and drawbacks of up-to-date SDN approaches that take energy efficiency into account. We also present a brief comparison of possible energy gain ratios of existing approaches, discussion on open issues, and a guideline for future research.

The rest of the paper is organized as follows. Section 2 presents background information on the definition, classification, and procedure of the SDN paradigm. Section 3 presents the importance of energy saving in SDN. Section 4 examines the principles, benefits and drawbacks of up-to-date SDN approaches in terms of energy efficiency. Section 5 compares the existing approaches from the view of energy gain, and finally Section 6 reports the existing issues, challenges and possible future research directions.