Software Defined Networking (SDN) is the moniker that the enterprise has accepted for the new breed of virtual network architectures abstracted from underlying hardware. But since software involves code, what we are really talking about is programmable networking. That raises a number of questions regarding the ultimate goal of this investment: programmed how? And programmed to do what, exactly?
For specialty chip developers like Xilinx, a key function will be workload acceleration. As the enterprise tries to satisfy the twin masters of increased data volumes and demand for real-time data services, the need to quickly navigate through an ever-shifting set of network resources will become paramount. To that end, the company has developed a number of programmable solutions aimed at increasing the flexibility of SDN and NFV (Network Functions Virtualization) environments. These include the Kintex UltraScale FPGA that enables the NVMe over Fabric architecture for hyperscale storage applications, as well as the SDaccel development environment optimized for standard programming languages like OpenCL, C and C++.
While these tools will benefit SDN, they are still programming the physical layer switch. But as F5 Networks noted in a recent white paper, few organizations have grasped the idea that programming the SDN layer is the only real way to fulfill the promises of a dynamic, vibrant network environment. To really make the most of your SDN investment, development should focus on four areas: automation and orchestration, application services management, resource provisioning and enforcement, and policy management. By addressing these core requirements, the enterprise will not only lower operational costs but reduce the risk of network failures and loss of service to users.
A programmable network is also more conducive to centralized control, says Accenture’s Amol Phadke. Rather than try to manipulate individual switches, SDN allows entire network pathways to be plotted automatically, which is a tremendous boon when it comes to the temporary, on-demand services and applications that are expected to populate enterprise workflows in the very near future. A programmable network is also the most efficient way – indeed, the only practical way – to enable the kind of machine-to-machine (M2M) communications that will make up a large part of the Big Data and IoT traffic that is expected to drive new insight and new levels of productivity over the next decade.
Still, most enterprises seem to miss this point when it comes to assessing the value of SDN. Simply separating the control plane from the data plane is only the first step toward a dynamic network, says Toronto-based tech consultant Andrea Knoblauch. The real action is in the APIs that enable network programming regardless of whether the underlying hardware has been decoupled from the network environment or not. And while centralized control and automation are worthwhile goals, the real prize is in network intelligence, which allows the environment to adjust and manipulate this programmable environment as operating parameters, workloads and user requirements demand.
Clearly, this level of functionality will not happen overnight. SDN deployment alone will likely unfold over a number of years, and then there is the training required to bring Dev/Ops up to speed.
But ultimately, SDN is set to remake the enterprise to a far greater degree than virtualization or even the cloud, because it fundamentally shifts the way in which data is gathered, disseminated and utilized – at once upping the value of data to the business model while lowering the cost of infrastructure.
It all depends on how well you program your networks.