SDN and Network Services
January 5, 2014 Leave a comment
Business value is created in two ways, by
- Lowering production and operations cost of a product or delivery cost of a service
- Creating new sources of revenues by bringing a new product or service to market
Expected impact of SDN on cost of network infrastructure and operations has received well-deserved attention. Another frequently mentioned aspect of SDN is ‘network programmability’ for new or customized behavior of the network – a big high-level idea that will have many different manifestations. In this entry, I will dig a little deeper and focus on the potential of network programmability to control resources to create new services and revenue from the network itself.
Network control over routing and bandwidth allocation is generally associated with reducing opex and capex – not revenue growth, which is the domain of innovative and usually over the top services that delight consumers. However, trend towards centralized control, an important element of SDN and PCE, presents an opportunity to bring background and often offline tasks, such as analytics driven decision support, path computation and restoration, closer to the subscribers and enable new network services.
A useful approach is to think through service assembly as a stack of technologies
- Applications running on devices or accessible through a browser
- Connectivity at the edge of the network
- Transport network
Applications running on top have been the real engine of growth, made possible by a confluence of complementary trends, convergence over IP, wireless connectivity enabling mobility, proliferation of hand-held devices and cloud computing. Value in network connectivity has been produced at the edge with use of policy servers for service differentiation and by extending Ethernet beyond local networks to metro and WAN. However, innovations in transport networks have been slow paced with a heavy reliance on physical infrastructure. There are some exceptions with pockets of innovation e.g. optical switching and remote provisioning at the DWDM layer but overall the key differentiation in the network infrastructure has been the size of the pipes – a commodity value proposition.
Separation of control and forwarding planes is at the center of SDN, which has the potential to change the stale state of innovations in network with a new approach to architecture and control. SDN has begun impacting the data centers and campus networks are expected to follow in near future. Based on efforts and initiatives by CSPs it seems the question for SDN, along with NFV, deployment in metro and WAN is ‘when’ not ‘if’ – 3, 5 or even 10 years in not very long for large infrastructure transformations.
Distributed control where routers make forwarding decisions independently and on limited link state information leads to unpredictable behavior because the sequence of network events alters which routing or forwarding decisions can be executed. SDN and PCE logically centralize control over forwarding decisions and treatment of traffic across the network. This reversal in control architecture and forwarding decisions presents the opportunity to make the network more flexible, responsive and predictable. For example, consider the following service features
- Capability to spin up and provision new network as a service is not a far-fetched pipedream; solution components to realize such capability already exist, more on that shortly. Conceptually, physical network infrastructure containing virtual networks is not much different from physical server as a container of virtual machines.
- Deterministic routing to restore services in an outage is a highly desirable feature that has remained elusive at layer 2 because order of events can change how, even if, a service is restored in an outage situation. At the physical layer restoration paths are deterministic but the process is fragmented and error prone, especially in mesh architectures.
- If DWDM is a good fit as a transport technology, benefits of SDN are clear. Unlike layer 2, optical networks are treated as point-to-point systems with little need for network wide end-to-end control. However, innovations in optical switching and remote provisioning have evolved the DWDM technology where network control will yield huge benefits.
Centralized fine-grain resource control sets the stage but to enable such network features, specialized decision support, analytics and infrastructure components are needed, for example
- Network topology and resource database
- Path computation capability that can apply resource constraints to calculation of paths
- Virtual switches that can share physical infrastructure
- Mechanisms for traffic separation among different virtual networks, such as the FlowVisor project at Stanford University.
Some of these enabling technologies are in production environments, some exist as prototypes and proofs of concepts but state of readiness of these enablers is not uniform in different types of networks, data center or access or core. One important point to note; although it is convenient to use the hypervisor and server virtualization analogy to conceptualize network virtualization, the latter is usually more complex. In non-trivial networks that have multiple switches and links, resources are distributed and there are vast numbers of possible allocations of link and switch capacity to different virtual networks – some allocation plans would be more optimal than others. This combinatorial optimization to allocate resources is not a facet of server virtualization but will be critical in commercial success of network virtualization.
In conclusion, the benefits of centralized control of network resources will create a more responsive and on-demand network infrastructure that will have a lower cost but also have capabilities to offer new network services, a source of new revenues.