Interest in transport software-defined networking has grown over the past year, but Roy Rubenstein finds there are still challenges to be overcome
Carriers are exploring a fundamental change to how they manage and operate their networks.
Distributed routing protocols that enable connectionless packet-based networking have long been the norm for carrier traffic. Now operators are exploring the idea of a central control instead of sprinkling intelligence across the equipment in their networks.
Having a global view promises operational benefits and more efficient usage of network resources. Software-defined networking (SDN) is at the core of the change – but, for it to be used, it must first be extended to the wide area network (WAN).
At its grandest, carriers view SDN as a way to optimise their complete infrastructure: network, storage, and processing. ‘The global orchestration of their compute, storage, and networking is the Holy Grail for the network providers,’ says Chris Liou, vice president for network strategy at Infinera.
SDN is still in its infancy but the technology is already proven in the data centre, where SDN enables search-engine and content service providers to cope with the demands of traffic flows created by virtualising servers and storage.
Extending OpenFlow for Transport SDN
The Open Networking Foundation (ONF), founded in 2011, is tasked with commercialising the OpenFlow specification for controlling switches that began life as an academic exercise.
The ONF has since set up the Optical Transport Working Group to define the OpenFlow extensions needed to control the optical layer, to create a multi-layer control plane.
‘OpenFlow, as it is now, is very packet-focused – Layer 3 and Layer 2,’ says Lyndon Ong, chair of the Optical Transport Working Group. ‘It was always a natural, technical extension to look at controlling Layer 1 and Layer 0.’
The Working Group delivers set work each quarter. The first quarter’s work concentrated on use cases and now under development is an architectural information model that includes the OpenFlow extensions.
Three use-cases have been chosen. An enterprise owning both the network and the data centre is the first. The second is network virtualisation, where a carrier provides connectivity for third-party data centres. Such networking must be allocated alongside existing services, requiring partitioning of resources across multiple users. The final use-case is not service-related, but addresses the broader topic of packet/optical integration.
The Working Group is pursuing two approaches for the transport extensions: the direct and abstract models.
Using a direct model, the SDN controller talks to each network element, controlling its forwarding behaviour and port characteristics. The abstract model controller, in contrast, may talk to a network element, or an intermediate controller or ‘mediation’. ‘This presents an interface as if you are talking to a bunch of network elements,’ says Ong. ‘But in fact it [mediation] is doing filtering and translation of the requests coming down [from the SDN controller].’
The direct model interests certain ONF members due to its potential to reduce the cost of networking equipment by moving much of the software from each element to the controller.
Factors favouring the abstract model include its suitability to use-case 2 where a client can partition resources and has some, but not complete, visibility into the network. Equally, the abstract model limits how much the SDN controller is exposed to the underlying network’s details.
Coriant has tackled this issue with its Intelligent Optical Control solution. The vendor has added a form of mediation – software that sits below the SDN controller – that provides planning and routing smarts across the transport network.
‘It can leverage the full performance of the WDM network because it knows all the planning rules,’ says Uwe Fischer, CTO at Coriant. ‘It does multi-layer optimisation across the transport layers.’ Coriant has added intelligence this way since it believes there are transport network-specific aspects that cannot be generalised easily.
The ONF Working Group expects its final model and extensions to support the direct and abstract models. Meanwhile, the work is to be completed by April 2014.