As networks evolve towards all IP, the ability to commoditize most network functions becomes economically appealing. To understand why the SDN may provide the greatest benefits at the metro network level requires an examination of the network bandwidth and economic challenges in the network core, metro, and access.
"Over-the-top" (OTT) video viewership is rapidly growing and putting great demands on existing network capacity. According to The Ethernet Alliance , network bandwidth is growing at a 59% rate annually. If the future IP network is to deliver virtually all video services, it must be capable of personalizing the video content.
Cable operators have two choices: install customer devices that mimic the functions of the Internet by performing such functions as caching and personalization in the home, or provide these as network functions. Network devices on customer premises will always be challenged by the inability to capitalize on ongoing economic improvements and difficulties in supporting evolving consumer devices, plus cost and deployment times. Velocity -- defined as how fast a provider can react to changing customer behaviors and economics -- will almost always be greater when more functionality is in the network rather than the customer premises.
A major goal of SDN is to both simplify and consolidate network functions, replacing specialized hardware with commodity switches, servers, etc. It enables vertical integration with application control over the network through SDN APIs. Utilizing SDN has the potential to improve service provider velocity and deliver better network economics. But where to implement SDN: the core; the metro edge; the access network?
Today's OTT viewing represents just a very small fraction of total overall consumer video viewing. If all broadcast and unicast video services must traverse the core, then it becomes a major bottleneck, and core capacity would need to expand by two orders or more. Core equipment is expensive, is highly risky to update, and has limited vendors whose equipment primarily uses specialized hardware rather than merchant chips. Thus, it doesn't make economic sense to use the core to deliver the personalized (unicast) services capability required in an all IP network, and therefore it is not the first candidate for SDN.
Access networks must deliver all services to/from residential and business customers. As video traffic explodes, access networks will need to accommodate all of this growth. Twisted pair, coaxial, and multiple fiber technologies are found in the access network, each with its own phy layer and connectivity requirements. Within fiber, there are multiple options such as EPON, GPON, switched Ethernet, and RFoG, for delivering residential and/or business services. It doesn't make sense to create SDN functions that are specialized for each access technology.
Between the core and access networks sits the metro network. 10G Ethernet transport and10G Ethernet switching are becoming cheaper by the minute. A wise sage said, "Never bet against Ethernet". Ethernet switches are simple and easy to make reliable even in outside plant installation, enabling low-cost extension of the metro network. Extending the metro using low-cost Ethernet transport and switching allows the access network to be much dumber and shorter, potentially driving down the cost of adding access capacity.
In this model, the metro becomes the place where the backbone terminates, provides local processing, and then, using switched Ethernet, connects over distances of tens of kilometers to the last few hundred meters of each specialized access network. It becomes logical to put the functions for IP services personalization at the metro network edge. Therefore, the metro network becomes the logical location for SDN functions.
Most network services can be software-defined utilizing commodity servers and switches. By using SDN in the extended metro, existing access networks can potentially become highly agnostic, with service functions instantiated to subscribers via SDN. Everything from DVR-like functionality to MEF-based services can be provided from the metro network.
There will likely be some service functions that require specialized hardware for many years. Examples include deep-packet classification and traffic-shaping functions (such as those that make up part of the cable network's DOCSIS specification), certain forms of encryption, etc. These specialized functions belong at the metro network edge as well.
As services delivery moves to all-IP, all services, whether residential or business, become IP streams, with their primary differences related to bandwidths, tolerable packet losses and latencies. Transport of business services and residential services over a single metro network becomes viable, despite very different QoS requirements. Network expansion costs can be minimized and service velocity maximized.
— John Holobinko, Independent Consultant