Today’s telecommunications networks are rather static and complex, involving equipment such as transmission nodes, routers, switches, and middleboxes, including firewalls, network address translators, and intrusion-detection systems. The IP packet traffic that flows from them—and where it goes—is controlled by management and control rules and policies. Management covers faults, configuration, accounting, performance, and security. Current networks are rather static and inflexible, and this makes it difficult to cope with future dynamic service demands and the need to quickly introduce innovative new technology.
Add to that advances in technology and the decreasing cost of IT and ultrabroadband connectivity as well as the Internet of Things, with its billions of new devices laden with sensors. And then there is the growing availability of relatively inexpensive open-source software and the falling price of hardware—both calling for networks to become more flexible and easier to upgrade.
Many experts, including IEEE’s, say much can be gained from the flexibility of software-defined networks (SDNs). The term refers to networks of equipment that separate the data plane (in which hardware does things such as forwarding IP packets toward their destination) from the control plane (where, for example, software is in charge of routing and traffic engineering). Moreover, in SDN the control software is not necessarily executed in the equipment but potentially in the cloud or in standard processing resources, such as IT servers.
A key aspect of SDNs is that they provide an array of abstractions and application programming interfaces (APIs) to program and control the functions and the services of network resources.
When talking about SDNs, virtualization is often mentioned. Virtualization is about creating logical resources, which are pieces of software running on a hardware host and emulating hardware capabilities, such as an x86 CPU (x86 is a family of backward-compatible, instruction-set architectures based on the workhorse Intel 8086 CPU). The novelty here is that the functions of most equipment, including the middleboxes, could be virtualized, incorporated, and moved as needed to various locations in the network.
SDNs and virtualization are not new concepts, but thanks to improved performance and lower costs, will soon help reinvent network and service architectures, experts say.
“Future networks will rely more and more on software—which will accelerate the pace of innovation,” says IEEE Member Antonio Manzalini, chair of the IEEE Software Defined Networks Initiative. SDNs give operators “the agility of a highly flexible and dynamic network, integrating IT and networking resources, and are capable of hooking together at the edge the huge number of terminals, machines, smart things, and even robots,” Manzalini says.
In fact, in time, the distinction between the network and what connects to it will disappear, says Manzalini, who is a senior manager at the Future Centre, part of Telecom Italia’s strategy and innovation department in Turin.
BUILDING A COMMUNITY
Much about the new architecture has yet to be defined. That’s why in May 2013 IEEE launched its Software Defined Networks Initiative, designed to explore how the technology will affect today’s networks. The initiative has formed committees on education, certification services, publications, standards, and conferences, and it is establishing connections with IEEE activities in related areas such as cloud computing, the Internet of Things, and consumer electronics.
“We want to create a large technical community of experts in academia and industry across the globe to work collaboratively to face the new challenges,” says IEEE Fellow Prosper Chemouil, chair of the SDN Conference Subcommittee. He is director of the Future Networks program at Orange Labs, in Paris.
To comprehend how SDNs work, it’s necessary to understand how traditional communications networks operate.
Each router and switch in a network is controlled at different layers so as to consistently address traffic transported as packets. Control planes decide how to route the traffic, which flows in the data plane, and where network applications are embedded. Much of the network’s intelligence for how it deals with traffic is scattered among all the devices distributed along the way.
The development of IP networks mainly relied on the distribution of intelligence in various network elements, and it proved to be robust in providing end-to-end services. The need for a logically central intelligence and control becomes clearer, however, when it comes to serving many types of applications with different quality-of-service requirements in a customized manner, Manzalini says.
That is where SDNs can make a difference. They can consolidate the control plane so that a single logically centralized software program controls multiple hardware elements. The control plane operates the state of the network’s data-plane elements (its routers, switches, and middleboxes) through well-defined APIs. It is this interface that enables applications to communicate with the infrastructure through an SDN controller, adds Chemouil.
“The network is configured and controlled through software, giving network providers more flexibility in how they use their resources,” he says. “In this sense, SDN controllers might be logically centralized, depending on the requirements of the network architecture.”
The future telecommunications network will look more like a distributed computational system, he continues. That means a network function could be executed either in the cloud or in the distributed telecommunications nodes.
“Software-defined networks are about creating very dynamic virtual networks out of a variety of aggregation nodes [routers with computing and storage capabilities], devices, and elements located at the edge of the network, down near the users,” Chemouil says.
The number of terminals, devices, and machines connected to telecommunications networks is growing exponentially. Fortunately, the principles of SDNs and virtualization are creating the conditions for reinventing network architectures to support the flexibility and dynamics of the growing number of advanced terminals and intelligent machines at the edge.
“The availability of open-source software and low-cost, high-performance IT hardware will allow service providers to install, program, and execute functions and services just like applications,” Manzalini says. “To the networks and the clouds, platform-hooking terminals will look like a fluid virtual environment of services.
“The border between cloud computing and networking infrastructure will start to blur,” he adds. “The border between the network and the terminals that are attached to the network will also disappear, because network functions could be executed in the cloud, in the network node, or even in terminals near the users.”
Network functions virtualization (NFV) is often paired with SDNs. The concept, which dates back to the 1960s, aims to use CPU virtualization and other cloud-computing technologies to migrate network functions from dedicated hardware to virtual machines running on general-purpose hardware. Virtualized network functions are appealing to network operators because they can be migrated and adapted to meet current demands and at the same time increase the utilization of network resources and decrease operating costs.
SDNs and NFV are mutually beneficial, but according to Manzalini, they do not depend on one another. For example, services can be developed in software and executed or emulated on logical resources without deploying an SDN, and vice versa. The combination can create an environment of virtual resources, interconnected by virtual links that are easily set up and torn down to serve multiple applications.
“Joining these two concepts can yield the most powerful network in terms of flexibility, programmability, and the ability to move virtual functions and resources while reducing costs and enabling new services,” Manzalini says.
The IEEE SDN Initiative will test experimental applications of software, such as network functions, services, capabilities, and hardware (nodes and systems architectures), he says, enabling SDNs and NFV to perform and conform to standards.
“The initiative’s work will accelerate SDN adoption by industry and help in creating new businesses,” Manzalini says. “SDNs will provide opportunities not only for IEEE but also for the academic and industrial worlds to explore.”