Matthias Buehner - Fotolia
We've had the promise of network revolutions before. IP convergence was a successful revolution, and the proclaimed frame relay/ATM revolution failed. So, it's tempting to look at virtualization, the cloud, software-defined networking and network functions virtualization as just four more candidates to be tested, but they add up to far more than that.
This time, we're not just talking about four new technologies, but about four faces of a network revolution that will shift the traditional view of services as being bound to the network infrastructure they run on to a service-independent infrastructure. In short, this could be the biggest change networking has ever faced and could lead to transporting services over virtual wires, rather than the physical layer in the Open Systems Interconnection, or OSI, model.
Services today are native behaviors of the devices we deploy in our network infrastructure, and that's a problem in terms of infrastructure optimization. To change the mix of services, you need to change the infrastructure, which may have an expected useful life of five to seven years. Then, if you change the mix of services or add new ones, you have to change your operations practices and retrain personnel. The tight binding of services to network resources means moves, adds and changes, which can be time-consuming and can generate massive disruption. These are the factors that virtualization, the cloud, software-defined networking (SDN) and network functions virtualization (NFV) aim to change.
Network services follow the first three layers of the old standard OSI model -- the physical, data-link and network layers. Still supporting these layers, the goal of the network of the future is to virtualize as many layers as possible to eliminate service-specific devices in favor of commodity white box or server elements.
To that end, SDN uses software control of forwarding to generate either Layer 2 data-link or Layer 3 network connectivity on white box devices, or hosted software switch or router instances. NFV deploys any network function, cloud feature or application element on cloud servers or customer premises equipment, and manages the collection of features to look like one cohesive service.
Virtual wires as physical layer replacement
We are already seeing SDN and NFV deployed in live services, so the question isn't whether this change is happening, but instead how fast and far it will go. The answer will depend on the first of the three OSI layers -- the physical layer. If the network revolution evolves to include virtual wires, then the infrastructure of the network of the future will be totally different. The term virtual wires replaced the notion of virtual circuits from ATM and frame relay when those technologies effectively went away.
All network connection services rely on moving bits around, and the core technology for that is optical networking. While optical pipes were originally both expensive and static -- remember SONET and SDH -- modern optics have driven down pricing radically, giving us wavelength-division multiplexing on fiber strands, and providing optical switching and grooming -- agile optics or reconfigurable optical add-drop multiplexer, or ROADM. Still, optical pipes are fat enough that it is not practical to dedicate them to users and services.
Virtualization applied above the optical layer to create virtual wires could solve that problem. By creating virtual physical-layer paths, virtual wires can be created as opposed to physical ones. Since any tunnel is a virtual wire, they already exist in some sense. SDN, however, could provide for electrical-layer virtual wires that don't require an Ethernet or IP network underneath them.
At the very least, this means networks could be partitioned at the virtual-wire level, and instances of software switches and routers could be added there to create Layer 2 and 3 services, such as virtual private networks. Although the dynamics of this shift are complex, if carried to an extreme conclusion, all services could be built that way -- even the Internet.
The transformative thing about virtual wires is they can be added and replaced with software processes. As packet layers themselves, they provide for the multiplexing or aggregation of traffic for efficient resource usage. Since route changes and aggregation are traditionally Layer 2 and 3 functions, virtual wires simplify the demands placed on these layers. Further, services and user traffic could be subdivided using virtual wires with no replication of expensive devices, which means you could build many services as independent networks.
Independent user or service networks of virtual wires could then be supplemented by NFV-deployed or cloud-resident virtual switches and routers to create services. VPNs and virtual LANs that are built today by segmenting switch and router infrastructure could be created by simply adding virtual devices to virtual wires. This would radically simplify deployment, changes and management. In fact, virtual wires could lead networking to a place where there are no switches or routers at all, just virtual wires built using SDN, and virtual switches and routers built on NFV. Such a network would be able to accommodate changes and respond to errors in a totally automated way, making it fully agile and extremely cost-effective.
Continue: In part 2, check out how NFV and SDN work together in network infrastructure design.
Where network hardware fits in the world of SDN and NFV
Check out the benefits of Central Office Re-architected for the central office
A look at NFV and SDN as effective business-level options for service providers