Lessons Learned

This survey experience has showed us several important points that require more attention from researchers to provide QoS in SDN networks.

QoS support for applications and service provisioning have been difficult tasks to achieve for quite a while even though newer applications such as video conferencing, VoIP etc. demand performance guarantees. Despite a large volume of work, QoS has not been completely deployed in today’s networks. A primary reason for this is the complexity of proposed QoS solutions and largely manual per-device config- uration of QoS knobs by network administrators. Supporting QoS for services and applications requires a well-defined automated QoS control and network management mechanisms in order to maintain the requested QoS performance over a network. A QoS control mechanism should provide an automated but fine-grained control for flow configurations. Also, it should be adaptive to dynamic workloads for dynamic QoS configurations based on network states. Furthermore, it should support legacy de- vices and large-scale networks like WANs. In addition, it should provide network-wide optimization in resource allocation by utilizing a global view of the network.

In recent years, some emerging applications, such as distance learning, video con- ferencing and so on, are becoming prevalent in networking world. Despite the advan- tages of these QoS-dependent applications for users, they still suffer from some issues regarding QoS or QoE requests of their users/customers. Firstly, today’s QoS based applications take into account only the network parameters as a QoS performance. However this approach does not reflect the user’s real satisfaction of provided services. Secondly, even if the user’s satisfaction, i.e. QoE, is provided, converting this QoE indicators to network-based QoS parameters is another issue. Also, this conversion

needs to be in a dynamic and optimized way. Thirdly, controlling and implementing QoS policies on the network is another issue for IPTV services.

An A-CPI enables applications to communicate with the controller to express their needs including dynamically specifying the QoS parameters of applications. Since they provide crucial tasks between applications and controller, network operators should consider certain points while designing A-CPIs. An A-CPI should be able to tolerate slow modifications of networks such as resource allocation for applications. It should also allow for determining the requirements beforehand using the application if possible. Defining different kinds of network parameters for different data types should be possible by an A-CPI. An interface should make sense for application developers while providing application metrics. A desired interface should not involve any application-related metrics such as response time. Instead, it should be able to convert these application-oriented metrics to network-based metrics such as delay, bandwidth etc.

3.13 Chapter Summary

Providing QoS is still a hot research problem in existing networking architectures. The emerging applications in the Internet (e.g. video streaming, VoIP etc.) generate diverse flows which require different treatments for each one. However, providing QoS needs of these flows is not easy with today’s networking models. Therefore, researchers has started exploiting the SDN paradigm and OpenFlow protocol since they bring centralized global network view, and more fine-granular flow management opportunities in networks. These features of SDN make it a better candidate in order to provide QoS for applications in easier and more flexible ways compared to traditional network architectures. This survey study has made a picture of QoS in OpenFlow-enabled SDN networks by surveying the current QoS-motivated studies in the field. It has organized the related studies according to the categories that are the most prominent ways in which QoS can benefit from the concept of SDN: Multimedia

flows routing mechanisms, inter-domain routing mechanisms, resource reservation mechanisms, queue management and scheduling mechanisms, Quality of Experience (QoE)-aware mechanisms, network monitoring mechanisms, and other QoS-centric mechanisms. It has also outlined the potential challenges and open problems that need to be addressed further for better and complete QoS abilities in SDN/OpenFlow networks and lessons learned during preparation of this survey chapter.

4 A SCALABLE HIERARCHIC SDN ARCHITECTURE 4.1 Abstract

All new networking architectures come with their own problems. Software De- fined Networking (SDN) has its own challenges which are needed to be addressed by researchers as well. One of the crucial problems with SDN is the control plane scalability since it is a bottleneck for its evolution. As the network grows, the num- ber of messages a controller receives also increases. This increase puts the controller scalability in the heart of problems of SDN. This chapter proposes a hierarchy-based network architecture along with an inter-AS routing approach with QoS. It exploits idea of levels in which networks with controllers reside and a main controller, which works like a broker, is on top of them to keep the global network state and view. The experiment results indicate that a controller in a hierarchic setting handles 50% less number of traffic than a controller in a non-hierarchic environment.

4.2 Introduction

Traditional networking is forcing its limits to meet the needs of today’s users, enterprises and carriers due to its limited capabilities. Configuration or installation of network devices and appliances requires more trained people and increases costs and take time to do so. Vendor dependency is an obstacle for network application developers and IT people to develop new types of network applications [1]. Increases in network applications, such as virtualization, cloud services as well as mobility and video content, requires more dynamic architectures of data centers, carriers or ISP networks.

SDN aims to handle above-mentioned drawbacks of today’s networking architec- tures. It brings the idea of separation of data plane and forwarding plane along with a controller to acquire the global view of the network. Network managers become more capable of efficient manipulating of network resources. SDN makes management of the network easy for network operators/administrators by providing flexible programma- bility resulted from decoupled forwarding and data planes. Network managers can easily manage their network resources by dynamic, automated and easy-to-handle applications. OpenFlow [34, 36] is the first standard protocol for communication of separated forwarding plane devices/applications (e.g. controller) and data plane dump devices (e.g. routers, switches). It removes the vendor-dependency of data plane devices and make them able to communicate with all kinds of controllers.

SDN is an evolving networking architecture and has not completed its evolution. Scalability, as in all new networking architectures, in SDN is one of the most im- portant challenges that will complete evolution of the SDN. As stated in [1], the scalability issue in SDN has not been focused by researchers as much as it deserves. Decoupling of data and forwarding planes is the most important reason to the scal- ability issue, particularly control plane scalability, since it requires management of data plane devices from a remote point (i.e. controller) and therefore control plane scalability becomes a focal point for the system. Also, as the number grows regard- ing the number of network devices such as routers, switches etc., the controller will need to handle more events and flow requests. This increase requires the control plane to be scalable with respect to the network size. In addition, the placement of the controller in a network has effect on the scalability of the control plane since the distance between controller and data plane devices introduces latency into the system [61]. There are some proposals to mitigate the control plane scalability issue of SDN in the literature. They are mainly categorized in either central controller- based [23,69] or distributed controller-based [66,68,72,82,83] solutions. Optimization techniques-oriented proposals are other types of the solutions. These proposals are discussed in more detail in Section 4.3. However, very few of them revolve around

a hierarchic controller-based solution. In this type of architecture, every domain has its own controller with their own network view and there is a Broker with a more global network view which orchestrates those controllers. This type of solution is more efficient and mitigates the issue of control plane scalability of an SDN network. That is the motivation for us to propose a hierarchy-based network architecture in this chapter.

In the remaining of the chapter, an extensive survey of studies, which aim to improve the scalability of the SDN with different techniques, in the literature is given in Section 4.3. Section 4.4 explains the details of the architecture and routing approach proposed along with its components. Section 4.5 clarifies how the proposed architecture works while Section 4.6 discusses the experiment results. Finally, Section 4.7 wraps the chapter up with concluding remarks.