Foundations of Modern Networking: Background and Motivation of Software-Defined Networks (SDN)

This chapter from Foundations of Modern Networking: SDN, NFV, QoE, IoT, and Cloud begins the discussion of software-defined networks (SDNs) by providing some background and motivation for the SDN approach.

• The requirements for a future all-digital-data distributed network which provides common user service for a wide range of users having different requirements is considered. The use of a standard format message block permits building relatively simple switching mechanisms using an adaptive store-and-forward routing policy to handle all forms of digital data including “real-time” voice. This network rapidly responds to changes in network status.
• —On Distributed Communications: Introduction to Distributed Communications Networks, Rand Report RM-3420-PR, Paul Baran, August 1964

Chapter Objectives: After studying this chapter, you should be able to

• Make a presentation justifying the position that traditional network architectures are inadequate for modern networking needs.
• List and explain the key requirements for an SDN architecture.
• Present an overview of an SDN architecture, to include explaining the significance of northbound and southbound APIs.
• Summarize the work being done on SDN and NFV standardization by various organizations.

This chapter begins the discussion of software-defined networks (SDNs) by providing some background and motivation for the SDN approach.

3.1 Evolving Network Requirements

A number of trends are driving network providers and users to reevaluate traditional approaches to network architecture. These trends can be grouped under the categories of demand, supply, and traffic patterns.

Demand Is Increasing

As was described in Chapter 2, “Requirements and Technology,” a number of trends are increasing the load on enterprise networks, the Internet, and other internets. Of particular note are the following:

• Cloud computing: There has been a dramatic shift by enterprises to both public and private cloud services.
• Big data: The processing of huge data sets requires massive parallel processing on thousands of servers, all of which require a degree of interconnection to each other. Therefore, there is a large and constantly growing demand for network capacity within the data canter.
• Mobile traffic: Employees are increasingly accessing enterprise network resources via mobile personal devices, such as smartphones, tablets, and notebooks. These devices support sophisticated apps that can consume and generate image and video traffic, placing new burdens on the enterprise network.
• The Internet of Things (IoT): Most “things” in the IoT generate modest traffic, although there are exceptions, such as surveillance video cameras. But the sheer number of such devices for some enterprises results in a significant load on the enterprise network.

Supply Is Increasing

As the demand on networks is rising, so is the capacity of network technologies to absorb rising loads. In terms of transmission technology, Chapter 1, “Elements of Modern Networking,” established that the key enterprise wired and wireless network technologies, Ethernet and Wi-Fi respectively, are well into the gigabits per second (Gbps) range. Similarly, 4G and 5G cellular networks provide greater capacity for mobile devices from remote employees who access the enterprise network via cellular networks rather than Wi-Fi.

The increase in the capacity of the network transmission technologies has been matched by an increase in the performance of network devices, such as LAN switches, routers, firewalls, intrusion detection system/intrusion prevention systems (IDS/IPS), and network monitoring and management systems. Year by year, these devices have larger, faster memories, enabling greater buffer capacity and faster buffer access, as well as faster processor speeds.

Traffic Patterns Are More Complex

If it were simply a matter of supply and demand, it would appear that today’s networks should be able to cope with today’s data traffic. But as traffic patterns have changed and become more complex, traditional enterprise network architectures are increasingly ill suited to the demand.

Until recently, and still common today, the typical enterprise network architecture consisted of a local or campus-wide tree structure of Ethernet switches with routers connecting large Ethernet LANs and connecting to the Internet and WAN facilities. This architecture is well suited to the client/server computing model that was at one time dominant in the enterprise environment. With this model, interaction, and therefore traffic, was mostly between one client and one server. In such an environment, networks could be laid out and configured with relatively static client and server locations and relatively predictable traffic volumes between clients and servers.

A number of developments have resulted in far more dynamic and complex traffic patterns within the enterprise data center, local and regional enterprise networks, and carrier networks. These include the following:

• Client/server applications typically access multiple databases and servers that must communicate with each other, generating “horizontal” traffic between servers as well as “vertical” traffic between servers and clients.
• Network convergence of voice, data, and video traffic creates unpredictable traffic patterns, often of large multimedia data transfers.
• Unified communications (UC) strategies involve heavy use of applications that trigger access to multiple servers.
• The heavy use of mobile devices, including personal bring your own device (BYOD) policies, results in user access to corporate content and applications from any device anywhere any time. As illustrated previously in Figure 2.6 in Chapter 2, this mobile traffic is becoming an increasingly significant fraction of enterprise network traffic.
• The widespread use of public clouds has shifted a significant amount of what previously had been local traffic onto WANs for many enterprises, resulting in increased and often very unpredictable loads on enterprise routers.
• The now-common practice of application and database server virtualization has significantly increased the number of hosts requiring high-volume network access and results in every-changing physical location of server resources.

Traditional Network Architectures are Inadequate

Even with the greater capacity of transmission schemes and the greater performance of network devices, traditional network architectures are increasingly inadequate in the face of the growing complexity, variability, and high volume of the imposed load. In addition, as quality of service (QoS) and quality of experience (QoE) requirements imposed on the network are expanded as a result of the variety of applications, the traffic load must be handled in an increasingly sophisticated and agile fashion.

The traditional internetworking approach is based on the TCP/IP protocol architecture. Three noteworthy characteristics of this approach are as follows:

• Two-level end system addressing

  TCP/IP protocol architecture

  The protocol architecture built around the TCP and IP protocols, consisting of five layers: physical, data link, network/Internet (usually IP), transport (usually TCP or UDP), and application.

• Routing based on destination
• Distributed, autonomous control

Let’s look at each of these characteristics in turn.

The traditional architecture relies heavily on the network interface identity. At the physical layer of the TCP/IP model, devices attached to networks are identified by hardware-based identifiers, such as Ethernet MAC addresses. At the internetworking level, including both the Internet and private internets, the architecture is a network of networks. Each attached device has a physical layer identifier recognized within its immediate network and a logical network identifier, its IP address, which provides global visibility.

The design of TCP/IP uses this addressing scheme to support the networking of autonomous networks, with distributed control. This architecture provides a high level of resilience and scales well in terms of adding new networks. Using IP and distributed routing protocols, routes can be discovered and used throughout an internet. Using transport-level protocols such as TCP, distributed and decentralized algorithms can be implemented to respond to congestion.

Traditionally, routing was based on each packet’s destination address. In this datagram approach, successive packets between a source and destination may follow different routes through the internet, as routers constantly seek to find the minimum-delay path for each individual packet. More recently, to satisfy QoS requirements, packets are often treated in terms of flows of packets. Packets associated with a given flow have defined QoS characteristics, which affect the routing for the entire flow.

Datagram

A packet that is treated independently of other packets for packet switching. A datagram carries information sufficient for routing from the source to the destination without the necessity of establishing a logical connection between the endpoints.

Packet

A unit of data sent across a network. A packet is a group of bits that includes data plus protocol control information. The term generally applies to protocol data units at the network layer.

Flow

A sequence of packets between a source and destination that are recognized by the network as related and are treated in a uniform fashion.

Packet switching

A method of transmitting messages through a communications network, in which long messages are subdivided into short packets. Each packet is passed from source to destination through intermediate nodes. At each node, the entire message is received, stored briefly, and then forwarded to the next node.

However, this distributed, autonomous approach developed when networks were predominantly static and end systems predominantly of fixed location. Based on these characteristics, the Open Networking Foundation (ONF) cites four general limitations of traditional network architectures [ONF12]:

• Static, complex architecture: To respond for demands such as differing levels of QoS, high and fluctuating traffic volumes, and security requirements, networking technology has grown more complex and difficult to manage. This has resulted in a number of independently defined protocols each of which addresses a portion of networking requirements. An example of the difficulty this presents is when devices are added or moved. The network management staff must use device-level management tools to make changes to configuration parameters in multiple switches, routers, firewalls, web authentication portals, and so on. The updates include changes to access control lists (ACLs), virtual LAN settings, QoS settings in numerous devices, and other protocol-related adjustments. Another example is the adjustment of QoS parameters to meet changing user requirements and traffic patterns. Manual procedures must be used to configure each vendor’s equipment on a per-application and even per-session basis.
• Inconsistent policies: To implement a network-wide security policy, staff may have to make configuration changes to thousands of devices and mechanisms. In a large network, when a new virtual machine is activated, it can take hours or even days to reconfigure ACLs across the entire network.
• Inability to scale: Demands on networks are growing rapidly, both in volume and variety. Adding more switches and transmission capacity, involving multiple vendor equipment, is difficult because of the complex, static nature of the network. One strategy enterprises have used is to oversubscribe network links based on predicted traffic patterns. But with the increased use of virtualization and the increasing variety of multimedia applications, traffic patterns are unpredictable.
• Vendor dependence: Given the nature of today’s traffic demands on networks, enterprises and carriers need to deploy new capabilities and services rapidly in response to changing business needs and user demands. A lack of open interfaces for network functions leaves the enterprises limited by the relatively slow product cycles of vendor equipment.