Disaggregation and reaggregation in optical interconnects: Where does each make sense?

Panelists – Nick Wright (LBNL), Gilad Shainer (Nvidia), Cyriel Minkenberg (Rockley), Chris O’Conor (Polatis/Huber+Suhner), and Bob Conner (X-Celeprint).

For the last ten years, we have enjoyed the fruits of disaggregation in optical interconnects. Before this trend began, optical interconnects were often bundled into much larger systems and integrated into their hardware, operating systems, and software. Thus changing to a new or faster optical link often meant replacing entire systems or products, and expensive ones at that.

Disaggregation, motivated partly by the advent of SDN, drove the modularity of optical interconnects so that as the optical technology evolved it could be swapped out without changing the whole system. Further, the adoption of Smart NICs accelerated the separation of packet processing and optical interconnects from both computing and communications devices. The result was an acceleration of optical performance and its more rapid adoption.

More recently we have seen the increasing reaggregation of optics into SoCs and other chipsets, mainly to save power and shoreline. But this comes with downsides, like having to replace a whole SoC or board just for a transceiver failure, not to mention a speed or other upgrade.

In this panel we will explore where disaggregation and reaggregation are exploited to offer overall benefits in system design, trading off speed, power consumption, reach, modularity, and cost.