Analysis of Difficulties in Operator SD-WAN Deployment

1. Introduction

Software-defined wide area network (SD-WAN) is an important branch of software defined networking (SDN) and is currently a hot area for SDN technology applications. The main goal of SD-WAN is to achieve the network quality of ordinary interconnection links to reach that of dedicated lines, reduce traffic costs, and improve bandwidth utilization by using a series of technical means such as software technology, application strategy deployment, and overlay networks; at the same time, for multiple sites, it can automatically select the best path according to the current network conditions and configuration strategies, achieve load balancing, and ensure the network quality of the entire network.

As a scenario application of SDN, the related technologies of SD-WAN are still evolving and maturing. Many domestic practitioners have proposed their own SD-WAN implementation plans. For operators, pipelines are the "business killer". How to introduce SD-WAN and make SD-WAN the whetstone of the killer has been the focus of research for operators in recent years.

This article will analyze the deployment of Google SDN, summarize the key factors for the successful deployment of its SD-WAN, and apply them to operator networks to help operators find the entry point for SD-WAN deployment.

2. Google Case Study

Google's network is divided into the internal network of the data center and the WAN. The WAN is composed of two backbone networks according to the traffic direction, namely the network interconnecting data centers (G-scale network) and the network accessed by Internet users (I-scale network). The Internet is used to connect Google's data centers around the world and is an internal network.

Google selected the interconnected networks between data centers and carried out SDN transformation. Through applications (software), it controlled the edge network devices of each data center to ensure that high-priority application traffic arrived with the lowest latency and low-level traffic filled the spare link bandwidth, thereby making full use of its expensive submarine optical cables.

Figure 1 Google B4 network

Google has achieved more results than many other SDN practitioners mainly due to the following reasons:

1. The network structure is simple and clear

There is a transformation using SDN. No matter how many data centers Google has, they are still limited, and the scale of WAN is simpler than that of carrier services. It is actually an internally controllable network with simple equipment, which is convenient for network transformation. By transforming the control plane and forwarding plane, the network is made smarter, the routing selection is more reasonable, and the network quality is better.

2. Strong R&D capabilities

Google is an integrated development and operation company. Its strong R&D capabilities or its own industrial ecosystem are sufficient to support the creation of a huge cloud-network integration system tailored for it from the underlying hardware and software.

3. Clear planning of business traffic flow and CoS level

In the traditional network era, Google has strictly differentiated its internal data by level, and clearly defined what kind of data is high priority, medium priority, and low priority. When implementing SDN transformation, the size of data traffic at each level must be clear, so that the overall traffic situation of each level can be summarized and the summarized CoS level planning can be specified.

4. Allow for a certain degree of error

In order to maintain high pipeline utilization, Google uses medium and low priority data to fill the pipeline while ensuring high priority data. When a failure or a burst of traffic occurs in a high priority business, low priority data will be delayed or discarded. At this time, it is necessary to allow low-level applications to make mistakes or require low-level applications to have sufficient tolerance.

Network intelligence, data classification and application tolerance are the key factors for Google's success in SD-WAN. Network intelligence is a capability, application data classification is the real core, and application tolerance is the selection standard. For users, "network transformation" of the carried applications to match them with the new network is an indispensable part of SD-WAN application.

3. Analysis of Difficulties in Operator SD-WAN Deployment

At present, the three major basic operators provide WAN services, and "BAT" is currently gradually implementing the work of building its own WAN. Taking China Telecom, the largest basic operator in China, as an example, the built WAN network includes the Internet network Chinanet, the high-quality professional network CN2, and various forms of MPLS/dedicated line networks deployed and constructed. From a hierarchical perspective, WAN can also be divided into access, metropolitan area and backbone networks. From the perspective of SD-WAN use, operators can be users of SD-WAN, hoping to provide better WAN services to "BAT" or directly provide SD-WAN services to customers through the deployment of SD-WAN; "BAT" will be potential deployers and users of SD-WAN, and they hope to build their own "black network" SD-WAN based on the operator's WAN network and bypass network operations. This section focuses on the factors of Google SD-WAN's success, combined with the actual situation of domestic operators' WAN networks, to analyze the entry difficulties of operators' SD-WAN deployment.

1. Difficulties in Intelligent Network Transformation

The core of network intelligence is to give the network a "smart brain" that can give the network different bearer forms and business forms according to various needs. At present, it is still difficult for operators to give the existing wide area network a smart brain and transform it into a network like Google B4 with the ability of "intelligent routing selection".

Taking China Telecom's WAN network as an example, the separation of the control plane and the bearer plane is implemented across the entire network, and the control plane is unified into the SDN Controller. This will involve more than 360,000 routers of various brands and models in the access network, metropolitan area network, and backbone network, as well as countless other network devices such as switches. How to coordinate the transformation of multiple brands and different batches of equipment according to one standard will cost a huge amount of money, which is unacceptable.

A more feasible solution is to transform the network through underlay, that is, to realize some "intelligent network" functions through unified network management. This solution can realize the flexibility of network configuration to a certain extent, have certain traffic scheduling capabilities, and improve business responsiveness. However, the biggest problem with this underlay method is that it limits "intelligence" to a limited range, which may be the same brand of equipment, a metropolitan area network or an operator. For WAN users, in recent years, as operators have increased their investment in basic networks, their basic network traffic carrying needs have been basically solved. The pipeline is already wide enough, and the quasi-intelligent network based on a city or a region cannot meet their needs.

Therefore, for a long time in the future, the operator's WAN network will still be a network lacking an overall brain and will not be able to meet the requirements of on-demand intelligence for the entire network.

2. Data classification and fault tolerance

The definition of data level is determined by the characteristics and requirements of the application. The same application has different requirements for network QoS in different scenarios. For example, in an application that uses active-active deployment, the active-active synchronization data requirements of the application are very high, while the daily system backup data can be relatively low.

For enterprises with multi-site interconnection needs, they rarely communicate with operators about system data classification. First, for existing systems, data classification involves the transformation of existing applications, and the existing enterprise dedicated line bandwidth price is sufficient to support its one-size-fits-all bearing method; second, data classification involves internal data management of the enterprise, and problems involving management are often not something that IT personnel can solve through design. Therefore, it can be seen that in reality, most enterprises are not as enthusiastic as they imagined about the "multi-path selection by business level" function of SD-WAN, mainly because enterprises cannot really make reasonable distinctions between application data levels.

The application data level is labeled at the server exit to let the network know the importance of application data, which is convenient for network hierarchical bearing and differentiated operation. It is the core of the operator's "cloud network integration" and the main driving force for the implementation of SD-WAN. However, since most business users have not yet realized the importance of data classification, it is difficult for operators to implement network transformation of typical SD-WAN business features such as "multi-path selection based on business level". For WAN operators, if users do not classify data, "application data will become customer data". Under the same price system, all customers are equally important to operators, and no mistakes are allowed. In this case, operators can only invest every year to expand the "pipeline bandwidth".

IV. Current Application Practices of SD-WAN

In reality, in the absence of network and application integration, operators can provide differentiated services by designing networks with different network bandwidths and OoS at network egress.

In most cases of WAN networking by operators, large enterprise customers will classify all applications into a higher-level network and adopt a network with high QoS guarantee, while small and medium-sized enterprises will choose a network with medium or low QoS guarantee. Operators will label data level of different customer levels at the network egress (such as in the form of VLAN tags) for differentiated bearer.

Before SDN, operators would connect enterprises to telecommunications rooms through optical fibers based on the above design, and implement "hard edge" equipment in the telecommunications room to carry different levels of customers in different levels. The problem is that customers need to bear the heavy enterprise professional rental price, and cannot flexibly change according to the business form; operators cannot flexibly select intelligent routes and fill traffic within the wide area network.

Combining the idea of ​​SDN and drawing on Google's SD-WAN design, at this stage, combined with the popularization and application of cloud computing virtualization technology, a simplified version of SD-WAN is introduced by softening the enterprise boundary and building a "soft edge", as shown in Figure 2.

Figure 2 Simplified SD-WAN

There are three types of dedicated lines in Figure 2. The red dotted line is an MPLS dedicated line with high QoS guarantee, the green dotted line is a virtual dedicated line provided by tunneling with higher QoS guarantee, and the yellow dotted line is an Internet dedicated line with lower QoS guarantee. All three methods are based on CPE devices under the control of the SD-WAN controller. The CPE device can be a white box device or a virtual machine. The SD-WAN controller can configure the routing method according to the designed data information and send the configuration to the CPE. The CPE selects different physical networks for the "most reasonable" data routing.

In this way, the internal network of IDC does not need to be transformed. It only needs to introduce the SDN concept at the network exit/entrance, which simplifies the implementation difficulty and avoids the problem of diverse large network equipment. First, through the SD-WAN controller, centralized management and automatic configuration of CPE equipment are realized, configuration management of various access methods is realized, network service quality is improved, and network flexibility and mobility are increased. Second, operators can use SD-WAN control to perform traffic filling operations to maximize network utilization and maximize network value. Third, providing network access interfaces for capable customers and opening large network routing scheduling to corporate customers in a conditional and limited manner can also allow small and medium-sized enterprises to save costs under tolerable conditions.

V. Conclusion

At present, most SD-WAN high-quality network solutions are actually an overlay solution carried on the original WAN network. By superimposing the network "tunnel" and taking advantage of the light-load characteristics of the high-quality network, the network quality is partially improved, which does not fully meet customers' expectations for SD-WAN. Based on the current status of the operator's network, this article analyzes the difficulties and limitations of transforming the WAN into SD-WAN in the process of transforming the existing operator's basic network to the Internet SDN. Combined with the analysis of the Google network, it proposes the entry point for operators to introduce SD-WAN at this stage to maximize their benefits.

This article discusses the difficulties in building SD-WAN for operators, focuses on analyzing the difficulties and feasibility of introducing SD-WAN in the entire network of operators at this stage, and proposes a deployment direction based on IDC applications. However, due to time and capacity limitations, further tracking and analysis of the actual application effects have not yet been done. I hope that I will continue to track the research of this project in my work, grasp the actual significance of SD-WAN for operator deployment, and make the operator SDN bigger and stronger.