Comprehending the capabilities of optics in the current data center networks.

The last few years have seen an exponential rise in data significantly owing to technological developments such as AI, cloud computing, IoT, and others. With this much growth of data, current data center networks are experiencing an exponential increase in network traffic. A data center typically consists of hundreds of thousands of connected servers to store everything people do online. Thus, as server-to-server traffic is increasing rapidly, it has become a challenge to meet the rising data storage and processing demand with existing technologies. Optical networks that use signals encoded in light to transmit information in different types of networks represent a promising alternative.

Optical network architecture has a lower CAPEX and OPEX when compared to traditional data center networks. It can transmit more than 10GB, 40GB, or 100GB/s over hundreds of meters or tens of kilometers relying on the technology used. Fiber optic systems are much thinner and lighter than copper ones and can be further bent. Such properties are beneficial when designing intra- and inter-rack connectivity, at which usually a large number of cables are bundled together. In data center networks, optical transceivers are implemented in pluggable modules, which can be easily plugged into switch racks and network interface controllers placed in servers.

Optics in Data Center

Optics is the only network connectivity cable that supports faster data rates, such as 50GB/s and beyond, in traditional network applications. This can be proved by referencing network switches, which contain the Application Specific Integrated Circuit (ASIC) designed to support 256 – 50GB/s switch ports.

As current data center networks are based on electronic packet switches, optical switching-based interconnect architectures have been proved effective in reducing or eliminating the electronic components for high-energy efficiency. Many researches have already demonstrated that optical solutions are able to minimize energy consumption significantly than electronic switching-based approaches. According to a report, the existing optical core switches for data centers can be alienated into two major categories: hybrid electronic/optical and optical switches. In hybrid switches, the electronic part addresses the fine granularity switching on the packet level, whereas the optical part is based on circuit switching offering high capacity.

Optical Shared Memory Supercomputer Interconnect System (OSMOSIS) is a high-performance optical packet switching architecture initiated by IBM and Corning in 2003. It is designed to develop the optical switching technology in supercomputers and resolve optical technical challenges of lessening the cost in High Performance Computing (HPC) field. This broadcast-and-select system is built upon wavelength and space division multiplexing, delivering low latency, high bandwidth, and cost-effective scalability.

Optical networks significantly present that storage needs not to be located in close proximity to processing, enabling data centers to become more modular, flexible, and distributed. They could be used to create large scale distributed learning. In recent years, the discussion around the optical data center has seen much uptake as the growth of data continues ascending. These data centers are advantageous in terms of fostering renewable energy sources and reducing energy consumptions.