In our modern society, people exchange an ever-increasing amount of data in digital form—both in their personal sphere (on-demand video, movie streaming and downloading) and in the professional domain (cloud storage and computing). Moreover, the ways in which this data are very diverse, from real-time low-rate voice over IP to high-volume communication between data centers.
Fulfilling this need relies on a sustained technological advance in data transmission and routing. The global communications network is required not only to handle high bit rates, but also to be agile so as to adapt in real time to its users' requests. The current “all over IP packets” model has been very successful in this respect to date, though new software techniques (notably SDN and network virtualization) need to be developed to manage the complexity of network requirements.
My view is that an all-software approach will not suffice in the medium term, even with progress in electronics. Large-scale packet switches are hitting barriers in capacity and energy consumption, and virtualizing the network may even exacerbate the problem by requiring more raw-metal performance.
For a long time, I and others have researched optical functionalities: given that information is mostly transmitted as light over optical fibers, it is logical to process it directly in the optical domain, rather than swamping overloaded optical-electrical converters. In particular, all-optical packet switches have been demonstrated; unfortunately, they are not drop-in replacements for electronic switches, notably they can't buffer packets and are thus much more vulnerable to contention. In general, it is not enough to deliver optical functionalities if they don't fit into a usable network paradigm.
Therefore, we need to either create a new network paradigm around these functionalities, or abstract away their nonidealities into a software layer. In both cases, SDN will be an enabler to manage this complexity.