One Hop Beyond: Creating Regional Connections From Data Centers Through Fiber Selection and High-Powered Amplifiers
By Wayne Pelouch and Leigh Frame
November 29, 2021
In recent times the way subsea cables interconnect the worldwide data network has changed. Historically submarine systems were terminated in cable landing stations sited as close to population centers and to the landing locations as possible. This reflected the origins of the data flows which were – if we go back far enough – circuits connecting users, and then more recently, user or enterprise-originated data paths, possibly to a data center or another user or enterprise. If you had been able to drill down and interrogate the majority of data flows, they would have been a transaction recognizable to a human, perhaps a phone call, perhaps a demand for information or streaming data in real time. But then, sometime in our recent lifetime, the world changed…
It is only 20 years since Google built its first data center (DC), roughly coinciding with the founding of Equinix out of Digital Equipment Corporation. It is roughly 10 years since Facebook’s first DC, and only 15 years since the first iPhone. That is a complete data revolution in 20 years. The consequence is that what fills the pipes now is mostly machine-generated data flows based on algorithms. Recent data suggests that between 50 and 65% of all internet traffic is ‘bots’. Good bots would be aggregator crawlers, search engine crawlers, social network bots, shop bots and monitoring bots, while the best known bad-bot would be a Distributed Denial-of-Service (DDoS) botnet. Unfortunately, research suggests that those bad bots account for perhaps 50% more traffic than the good guys. These patterns of data traffic have led to industry and market revolutions – compare the customer base and funding sources for subsea cable systems today to twenty years ago, and to an imperative for technology to keep up!
When subsea cables connected people, the signals arrived at the cable landing station and from here the signal was regenerated and transmitted terrestrially inland towards PoPs and then DCs based in cities and hubs. Now that machines spend much of their time and energy (literally) exchanging with each other systems are primarily designed to connect Data Centers to Data Centers, directly facilitating an end-to-end network that as far as possible resembles one pane of glass. Those Data Centers are also in locations where proximity to people is almost secondary, compared to energy cost, green considerations, land cost, legal and physical security and other factors, which may dominate.
As the number of new trans-oceanic systems accumulates, they increasingly connect seemingly random places optimized around machines, geo-politics and perhaps the environment; regional connectivity needs to be added and improved to support onward transmission of critical data, both in-country and to Data Centers in neighboring countries. Decommissioning of existing multi-landing trans-oceanic systems, such as TAT-14 or CANTAT-3 results in a reduction of local links extending from regional hubs. Similarly, a number of existing unrepeatered links are approaching their end of life and can be replaced by much higher fiber count unrepeatered or SDM (Space Division Multiplexing) systems, utilizing modern higher quality fiber.
High optical performance and transmission capacity are achieved subsea because the newly deployed subsea fibers are low attenuation and often high effective area, and the amplifier spacing is at regular and relatively short intervals. However, this is typically not the case in terrestrial fiber links and in unrepeatered subsea links (~ 200 to 500 km) that do not have electrically powered subsea amplifiers. Links that fall within these categories are often more challenging and result in a bottleneck of transmission capacity. There is also a trend to extend the higher performance subsea links through these lower performance sections to avoid regeneration costs at a subsea cable landing station. Methods to improve the performance of these bottleneck sections are discussed below.