Software Defined Networking (SDN) has been arguably termed as a revolution in the networking industry, but it is certainly not observed to be so. The promise of replacing the existing/traditional networking devices with ‘generic white box programmable hardware’ was, unfortunately, never fulfilled and the SDN community no longer envisions that SDN will be used to move packets over the network. Many factors contributed to slow development and deployment of SDN, the key reasons being the fragmented and convoluted software ecosystem around SDN, and the failure to standardize communication protocols between the hardware and the software.
"Data is the new oil and SDN is helping extract more oil"
Now looking at the positive side, SDN has been a boom to the NFV world, enabling a highly agile and programmable architecture for high-performance specialized network functions. SDN has also assisted in the advancement of IOT, Artificial Intelligence (AI) and Network Security in the networking world.
Let us look at the three key benefits SDN offers and how they impact the aforementioned technologies.
Match-Action Pipeline: The very first idea and meaning of SDN was to have programmable match-action data-plane, wherein the generic white box hardware is instructed (by an SDN controller or an application) via a protocol (such as Openflow or P4) to perform certain dynamic/reactive matches and actions. This paradigm offers flexible control over network flows and allows an application to manipulate flows/packets in the way it wants. This fundamental idea behind SDN allows IOT, AI and Security applications to implement flexible user-defined or machine-defined policies and aids their sophisticated functions.
Software/Hardware Disaggregation: The separation of the control plane and the data plane lets manage and balance network functions between software and hardware. CPU hungry sophisticated tasks can be processed in the software (at slower speeds), while repetitive and primitive tasks can be processed in the hardware (at much faster speeds). This separation is helping VNFs (including firewalls, load balancers, deep packet inspection) achieve unprecedented scale. One interesting use case would be: An AI application only needs to see a fraction of traffic to make certain inferences and post which traffic/flows can be transferred to be processed in the hardware as instructed by the application. Also, fast-growing network security applications (which require frequent security updates and signatures) will benefit from disaggregation due to the ability to rapidly innovate and update security functions without touching the hardware appliance.
Telemetry/Data Exports: Though Openflow suffered limitations on how fast an application can export streaming telemetry data from the switch, the onset of P4 (combined with Barefoot Networks Tofino chip) has opened up promising possibilities. Exporting streaming network data at ultra-high speeds would enable new evolution of data-hungry Machine Learning applications, particularly in network security and automation (self-driving networks). This is also highly impacting the IOT industry and is being used to enhance IOT security, to lock down devices and to enable data-driven decisions. As often quoted these days, ‘Data is the new oil’,, and SDN is helping extract more oil.
Though many have argued and confused the meaning of true SDN, these three features are fundamentals of SDN and should be leveraged by the networking community. Most of the network operators (including ISPs and IXP operators) have been frustratingly slow in adopting what SDN can offer, yet Google stands as a true inspiration in harvesting the power of SDN.