Discussion on the Application of SDN in Wide Area Network

  Internet industry application trends and problems facing wide area networks

For a long time, the services provided by WAN have always had problems such as long service provision cycle and inflexible service dynamic adjustment, which not only seriously affects the service application experience of users, but also restricts the rapid development of operators' services. This problem is particularly prominent in the mobile Internet era where service traffic is increasing rapidly and the demand for WAN services is changing frequently.

With the rapid development of the Internet, especially mobile Internet, business applications and user scale based on the Internet and mobile Internet have shown an explosive growth trend. The Internet industry presents the following trends: mobilization of business applications, cloud-based platforms, explosive growth in video services, and rapid spread of OTT business applications.

Digitalization of corporate business has become a trend, corporate IT systems are gradually migrating to the cloud, and corporate mobile office is becoming popular; ICT service providers are building cloud data centers to provide various cloud computing business services to major companies. The link access of cloud data centers requires wide coverage, high bandwidth, low latency, high quality, and agile wide area network support.

The bandwidth requirements for cloud access and cloud interconnection of public, private and hybrid clouds change frequently, requiring WAN services to be more agile. The traffic patterns of various cloud-based mobile business applications, big data, and especially the rapidly developing IoT business, change rapidly and cannot be predicted in advance. The imbalance of network traffic is more serious, requiring more agile WAN operations. Cloud applications in various industries are rapidly innovating and going online quickly. The traditional manual configuration mode cannot meet the agile innovation and online speed of cloud applications at all. It is necessary to open APIs for WAN and develop and operate on demand.

These industry application trends have put forward higher requirements on the architecture, bandwidth, performance and operation of the WAN. In order to adapt to the new demands of the new trend of Internet services in the cloud era for the transformation of the business model and operation model of the WAN, the introduction of SDN in the WAN has become an important means of optimizing the WAN.

Characteristics and architecture of SDN

As a new network architecture technology, SDN essentially separates control and forwarding, and centralizes and intelligentizes control and management. Its main features are as follows:

1. Separation of forwarding and control

Forwarding is separated from control, and software is separated from hardware. Distributed control is changed to relatively centralized control of multiple controllers. Open standard interfaces are used between layers, and general hardware is used. The hardware only completes necessary forwarding.

2. Network intelligence and centralized management

Centralized management and scheduling of routing calculation, business logic, data forwarding, etc. Implementing software-based control strategies is conducive to network intelligence, automation and hardware standardization.

3. Network Programmability

The network gradually acquires open programmable capabilities, enabling automated management, making the network more flexible and scalable.

4. Software-based business customization is conducive to the testing and rapid deployment of new businesses.

5. Relatively independent of NFV technology but also mutually coordinated, virtual network functions are called and opened as business capabilities.

SDN architecture widely recognized by the industry

The model architecture is divided into three layers. The infrastructure layer is mainly composed of SDN switches that support the OpenFlow protocol. The control layer mainly includes the OpenFlow controller and the network operation system (NOS). The controller is a platform that can directly communicate with switches using the OpenFlow protocol; upward, it provides an open interface for application layer software to detect network status and issue control policies. The application layer at the top level is composed of many application software that can execute specific control algorithms based on the network information provided by the controller. The results are converted into flow control commands through the controller and sent to the actual devices in the infrastructure layer.

Problems affecting the deployment of SDN technology in WAN

1. Security issues of centralized control

The controller, as the core of the SDN network, will become the focus of attacks, and it is necessary to solve the problems of SDN controller anti-attack and redundant backup. In the scenario where the forwarding device is controlled by multiple controllers, there may be risks such as controller impersonation and illegal abuse of forwarding device resources.

2. Performance issues of controller and forwarding plane

The Openflow protocol requires specialized chip support. General-purpose chips have poor performance, customized chips are expensive, and are difficult to apply using SDN technology.

3. End-to-end control coordination issues

The east-west interface has not yet begun to be standardized, and it is difficult to make controllers from different manufacturers universal. The current collaborative orchestration layer only solves the problem of issuing business policies, and end-to-end topology discovery and path calculation are difficult to achieve.

Analysis of SDN application scenarios in wide area networks

First, traffic balancing can be achieved between multiple links in the WAN.

On the WAN, traffic imbalance often occurs due to uneven user distribution and different concurrency characteristics of various services. The destination of the WAN link is usually the metropolitan area network or cloud IDC, and there are huge differences in the user scale and number of services. Traffic imbalance is prone to occur on several links. In severe cases, some links are close to congestion while the rest are still lightly loaded. This situation often occurs on the interconnection links between the primary trunk line, secondary trunk line and cloud IDC of the transmission network.

Unbalanced link traffic involves many complex issues such as routing, specific business application layer, and link quality. At the same time, in the era of rapid popularization and application of cloud computing, big data, and mobile Internet, the instantaneous dynamic changes of traffic are extremely fast, and manual intervention is not feasible. Therefore, it has always been a difficult problem that plagues operators' construction and operation and maintenance technicians.

Introducing SDN intelligent manager on the WAN, using SDN intelligent global control capabilities to collect real-time status information of each link in the WAN to monitor network status, calculate link utilization and bandwidth usage, dynamically obtain network status from switches and routers, calculate the best, collision-free path for data packets, and dynamically adjust routing to avoid network congestion, thereby achieving load balancing of WAN transmission link traffic. At the same time, SDN intelligent management can also be used to dynamically change bandwidth and accelerate the transmission of WAN link data.

Secondly, SDN can optimize the utilization of WAN links.

Currently, the path and channel network bandwidth is predicted based on the end-to-end network peak of the communication. The capacity of switches and routers is statically fixed, and the routes are reorganized based on the failure of the router uplink. The path selection will be repeated, and resources will compete, resulting in retries and failures again. The network utilization rate is usually less than 30%.

After SDN improvements, with a global view of network requirements and network organization, the best and most clear path can be calculated for all traffic, and the network behavior of data packets can be determined, thereby reducing the need for protective bandwidth. Once a failure occurs, it can be changed to another pre-calculated network path based on the error guidance. Higher network utilization and faster network convergence require only a small amount of resources to obtain higher network quality, and network utilization can be stabilized at more than 70%.

In the backbone network, there are many businesses that need to dynamically provide network capabilities on demand. For example, the bandwidth required between cloud data centers is usually not too large to support daily business operations. However, when there are businesses such as virtual machine migration and storage data replication, a large bandwidth is required to meet business needs. The network traffic bandwidth between two IDCs needs to be dynamically adjusted over time.

The traditional model is to design a fixed rate network bandwidth based on the peak value of end-to-end traffic between IDCs, which leads to seriously low bandwidth utilization; after the introduction of SDN, by adding OpenFlow-based switches to the grid intersection nodes, the cloud service platform can redistribute the network traffic bandwidth capacity anytime and anywhere according to demand through the SDN control layer, thereby ensuring the performance of network bandwidth. Compared with the traditional reserved space method, the new model saves about 65% of capacity for key communication transactions.

Multi-level network management can be achieved in the wide area network. Before SDN was adopted, the IP network was separated from the transmission network, and the network needed to be managed through different tools and different technical combinations. After SDN was adopted, the traffic was controlled with unified concentration and optimization as the driving force, so that the traffic flow on the network was always just right; network transmission became dynamic, mainly interacting through IP; SDN would selectively cut in to the optimization of network traffic, increase bandwidth and improve performance by re-coordinating and connecting new network ports, and reduce the impact of traditional network cutover.

Finally, SDN can effectively improve the service quality of the wide area network

Currently, the service content, application system, and service layer of the wide area network are tightly coupled with the network. The networks in different fields such as mobile networks, WiFi networks, and fixed broadband are also tightly coupled, resulting in low resource utilization or poor performance.

After SDN improvement, the control layer has a global view of the network organization and can provide this information to the application service layer. At the same time, it can select the best link between the user end and the IDC and specific servers. Only a small amount of resources is needed to obtain higher service quality, aggregate multiple technologies to provide services, and optimize performance in multiple aspects such as QoS, energy consumption, and utilization.

In the application scenario of server load balancing in the perception network, the current server load balancing method is to select servers from the server cluster to share the load based on the availability of the servers and the current resource utilization, without considering network factors. This will cause the selected servers to be idle, but the network path is congested. After using SDN, applications that perceive network loads and SDN controllers can select a network path with sufficient capacity to achieve end-to-end load balancing. This scenario is particularly suitable for latency-sensitive services, enabling them to obtain better service response time and effectively improve the user's service experience.

At present, it is not realistic to fully deploy SDN in the WAN. The SDN standards have not yet been unified, and there is a lack of systematic standardization for the needs of the WAN. The solutions of mainstream manufacturers that can be used vary greatly, there are many private protocols, and cross-manufacturer interoperability is difficult. At this stage, we can only choose SDN controllers from specific manufacturers based on specific needs. With the continuous improvement of SDN technology, the introduction of SDN technology applications in the WAN will be fully launched, and the performance of the WAN will definitely be well optimized, which will provide better services for the rapidly developing cloud computing, big data, Internet of Things and other business applications.