Cédric Ware, Mounia Lourdiane
Despite virtually all data being carried by networks in the form of
optical signals, in conventional networks with a layered paradigm,
the use of optics is confined to the physical layer. Most switching
and all processing is performed by electronics, requiring numerous
optical-to-electronic conversions that create bottlenecks in data rate,
latency, and energy consumption. Optical functionalities can perform
similar functions, though not equivalent, which precludes their
deployment even after decades of research.
Going beyond the layered-network paradigm would not only allow a better,
cross-layer optimization of the network; it would also enable a smarter
use of optics. The challenge in such a paradigm change is redefining
the abstractions in which networks are expressed. A new “cross-layer”
model, likely built on the SDN ecosystem, would take into account both
optical functions’ full benefits and constraints, encapsulating their
non-idealities whenever convenient—just like managing a complex
software problem by pushing the complexity into small units with
generic interfaces such as device drivers or object-oriented
frameworks.
It is thus our belief that developing a network simulator generic
enough to handle optical functionalities will lead to relevant
software-based abstractions. We have developed a number of specific
simulators that allowed us to quantify network-level performance gains
brought by a single functionality at a time: optical packet switching,
all-optical wavelength converstion, quantum-key distribution. We are
now moving to level up in genericity and modularity. When the
simulator or simulation framework has the flexibility to represent
all possible optical functionalities, we should have found suitable
abstractions to represent and manage a cross-layer optical network.