No Jitter is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

How Software Defined Networking Will Change Communications

The huge, amorphous thing that we call "The Net" has become a communications ball and chain. The routers and switches at the core are proprietary, monolithic boxes: they're expensive; the vendors control the standardization bodies and they determine the pace of innovation; it's slow. Although these routers and switches employ standard IP and Ethernet as distributed-routing protocols, they do not expose any internal programming interfaces or enable access to the forwarding plane.

All of this means that the core infrastructure hasn't changed that much in the last 40 years--the real innovation that drove the Internet boom took place at the edges. In that time corporate computing and communications environments have evolved continuously, and in recent years the pace of innovation has accelerated. Even low-cost commodity servers are yesterday's hardware. Today they're a virtual, pooled resource that operates very efficiently in private and public clouds. IT is in control of these resources, but when it comes to the Net, users get what they get. There is an urgent need for new standards that will address the problems of scale and flexibility. That need cannot be met by applying more Band-Aids.

The OpenFlow Protocol
The OpenFlow protocol that Nick McKeown and colleagues developed at Stanford went back to networking basics. The key objective was to enable third-party, e.g., enterprise IT, control of Internet routing, and it would be realized by allowing the path of packets to be determined by software running on a separate server. In a nutshell, OpenFlow enables control--sophisticated traffic management control--by providing access to the forwarding plane of a network switch or router, i.e. control is separated from forwarding. This is a groundbreaking development: the key that unlocks the potential of software-defined networking.


Figure 1 illustrates how OpenFlow provides an open interface to packet switching. And yes, it really is that simple. In addition, the network operating system moves out of the mainframe-type boxes and into a logically centralized place, and this provides a global view of the network. The features also move out of the boxes, thereby allowing users to create new features themselves. No longer are they dependent on the vendors.

The Open Networking Foundation (ONF) was formed in order to take this development forward into the real world. ONF is a standards-setting group that recognizes the way cloud computing has blurred the distinctions between computers and networks, and that this process needs to be speeded up through simple software changes in telecommunications networks. Dan Pitt, Executive Director at the Open Networking Foundation, said: "Networking has got to get closer to the computing environment and ideally it would become an integral part of the corporate ICT infrastructure. Moreover, developments have to be spearheaded by the users’ requirements and not those of the vendors."

Now It Gets Interesting
In Figure 2 we see how the hardware has been virtualized, just as app and storage servers are in cloud computing. This allows those large, expensive mainframe-type boxes to be replaced with much simpler, packet-forwarding devices, e.g. regular Ethernet switches, which are a commodity product.

It's clear that replacing all the kit that is out there in the Internet will take time and vendor agendas will kick in, but what makes this second-stage development really interesting is the ability for enterprises to create their own private, software-defined network. For example, it would seem to be an ideal way to connect data centers. Not only is there no reliance on vendors, now there's no reliance on the Internet. Instead, companies such as Amazon and Microsoft could also use it to create a public cloud.


Figure 2 shows how the software-defined network model is close to that of the layered, virtualized architecture of mainstream computing. The fact that there are different control programs at the top of the stack indicates that the architecture is open, i.e. the owner of the private network can implement whatever functionality they want. They do not have to run back to the box vendors or the standards committees. The network is configured to match the requirements of the applications. The slicing layer, FlowVisor, developed by Stanford allows separate, secure virtual networks on the same physical infrastructure.

Over the years, we've become accustomed to an architecture in which every node has its own control program, which it uses to send information to lots of other nodes. It was a great way to build a seriously robust (bomb-proof) network, but that constraint is no longer required. That said, if a link goes down in a software-defined network then traffic can be rerouted because a mesh architecture is employed; and because the Ethernet switches are a commodity, it's affordable. In addition, mesh architectures provide load sharing: it's an intrinsic feature.

The functionality of these devices is relatively limited: they don't need to make any local decisions. The network operating system knows the topology; it works out the routes and gives instructions (commands) to the switches, which they put in their forwarding table. The switches do a blind look-up and take the relevant action. A blind lookup means they don’t need to understand what they are doing. OpenFlow sends the commands, and the control program computes the routes. They can include policy, security, traffic engineering, regulatory compliance and so on.

There are some interesting implications in this model. In addition to handling heavy-duty traffic between data centers, the network can be employed for all internal enterprise communications, including voice, which is just another data type. Also interesting is the possibility that communications between networks will occur as a result of industry consolidation around a number of interfaces that are determined by the marketplace and not by a standards committee. That is the way things get done in the computing space. And the icing on the private network cake comes from the fact that OpenFlow enables coexistence with the company's legacy network, which in turn means that it can be deployed incrementally, one OpenFlow switch at a time. No need for a forklift upgrade of the entire network.

This development can be deployed in both peered and non-peered networks. A non-peered network would typically be an enterprise or service provider with network islands or regions, as with local data centers, WAN interconnects, access versus core networks, wireless backhaul networks, and so on. In this case the overall owner could organize the control function in different ways.

For links between different network owners, e.g. a carrier and an enterprise, SDN is an ideal way for the carrier to offer virtual control to the enterprise of its virtual resources at the carrier. This would be through suitable service interfaces in the orchestration layers above OpenFlow. This is something that the ONF is starting to study. Between carriers, with actual peering, the equivalent to peer service interfaces is required. This is another area that the ONF can investigate, as the foundation of OpenFlow is quite solid.

Conclusions
Today, enterprise computing is conducted in a very efficient, dynamic and flexible environment, but it's constrained by rigid communications concepts and networks that were designed 40 years ago. Software-defined networking does more than change the rules of the network game; it throws them out the window. Now network functionality is determined by users and managed by the IT, not kit vendors. There is increased awareness that SDN represents the future for networking.

The ONF expects to see carrier announcements later this year and some enterprises are leading-edge adopters. The heavy hitters on the ONF Board (see the sidebar below), with their deployment of huge data centers, are obvious candidates. However, many enterprises will exploit SDN transparently through carrier services.

SIDEBAR: More on the Open Networking Foundation
The ONF's goals include:

* Advancement of the SDN (including standardization); but it is not a regular standardization organization like the ITU
* Make software-defined networking the new norm
* Foster a vibrant market for SDN products, services, applications and users
* Create the most relevant standards in record time to support a switching ecosystem based on the OpenFlow Protocol
* Be driven by users and user needs. Board members include Google, Facebook, Yahoo, Deutsche Telekom, Verizon, NTT Communications, and Microsoft. No kit vendors
* Standardize as little as necessary, as opposed to standardizing as much as possible