SYSTEM CAPACITY SCALING: DATA CENTER MODEL
In order to better understand these trends it is important
to see where capacity is most being used and where
bottlenecks are occurring. Data centers are an excellent
example of an application in which parallelism has been
used with optics, which are optimized for efficiency and
size. In a recent study, a team at the University of Arizona
examined the aggregate capacity of data center networks
on different length scales (shown above). The figure shows
three capacity peaks: 1) within the processor, 2) inside the
data center, and 3) national scale networks. Two bottlenecks
between these peaks are related to the intra- and inter-data
center interconnection networks—two focal areas of CIAN
research. Optical interconnects predominate above 1m
distances, but through photonic integration promise more
capacity inside the servers. At the same time, the reach and
programmability of data center networks are being extended
outside the data center to close the bottleneck.
D. C. Kilper, H. Rastegarfar, “Energy challenges in optical access and aggregation networks”
Phil. Trans. R. Soc. A 2016 374 20140435; DOI: 10.1098/rsta.2014.0435 (2016).