What are the best practices for designing and deploying extremely high density data center interconnections?

May 20, 2020

Leave a message

Data center campus layout


A quick Internet search for announcements of outsize or multi-tenant data center spending can turn up multiple expansion plans totaling billions of dollars. What do you get out of this investment? Usually, it is a data center campus, which consists of several data room modules located in different buildings. These data rooms are usually larger than a football field, and the flow between the data rooms is usually more than 100Tbps.


There are many detailed reasons why these data centers have grown so large, but we can simplify them to two trends. The first is the exponential increase in east-west traffic from machine-to-machine communications. The second trend is the application of flatter network architectures such as crested and Clos networks. The goal is to build a large network structure within the campus that will enable data transfer between data centers to reach or exceed 100Tbps.


It is conceivable that a network of this scale will encounter a number of special challenges throughout the network, from power and cooling to the connection of devices. Various methods have been evaluated to provide 100Tbps(or even higher) transmission rates on network device interconnections, but the common model is to transmit at a lower rate through multi-core single-mode fibers. It is important to note that the length of these connections is usually 2-3 km or less. Through our modeling analysis, using more fiber to transmit at a low data rate remains the most cost-effective approach, at least for the next few years. This cost model reveals why the industry is spending so much money to develop high-core cables and associated hardware.


Now that we understand where the demand is, we can turn our attention to alternatives in the data center interconnection market. The industry agreed that the ribbon cable was the only viable solution for this application. The traditional loose tube optical cable and single-core optical fiber end connection installation time is too long; the optical fiber joint fusion hardware is too large and not practical. For example, a 3456 fiber cable with a loose casing design takes more than 200 hours to finish the fusion, assuming each fusion takes four minutes. If you use the ribbon configuration, the weld time drops to less than 40 hours. In addition to saving that time, the capacity of a ribbon splice is usually four to five times the density of a single-core fiber splice at the same hardware footprint.


Once the industry has decided that the ribbon cable is the best choice, it will soon become clear that the traditional ribbon design cannot achieve the required fiber density in the existing pipeline space. Therefore, the industry set out to double the density of optical fiber inside the traditional ribbon cable.


Structure of optical cable


There are two ways to design the structure of the cable. The first method USES a standard matrix strip with a more tightly encapsulated subunit, while the other USES a standard cable structural design with a central or slotted design and a loosely coupled ribbon fiber design that can be overlapped.



Send Inquiry