The edge: open ended

Editorial Type: Feature Date: 03-2020 Views: 882 Tags: Networking, Hyperscaling, 5G, Nexans
Some IT is moving to the network edge. Mike Connaughton, Technology Application Consultant at Nexans explains how it might unfold

For some, the term edge may seem like a new one, but it's actually the result of decades of technology evolution. Ever since computers first arrived there has been constant tension between centralised and distributed processing which changes with technologies and user requirements and ushers in new players and the demise of established ones. Presently there is a centralised versus distributed processing ebb and flow through the relationship between edge and hyperscale Data Centre's (DCs).

During the eighties processing took place on mainframes with user access from workstations. This sacrificed the flexibility individual users had for the efficiency of significant centralised number crunching and storage power. The 90s saw processing shift to powerful desktop PCs where software packages provided greater configuration freedom which often caused system conflicts. This prompted a move back to more centralised control, sometimes limiting user freedom and ingenuity. Since around 2010 large numbers of users have worked with cloud-based files and applications from any location, at any time, providing network latency permits.

Three types of companies are striving for dominance, thereby shaping the edge in different ways. Telco providers, with their historic end user connection have significant amounts of cable in the ground as well as numerous distributed central offices. These assets can be utilised for edge DCs relatively easily. The installed base outweighs disadvantages related to regulatory hurdles and sunken costs.

A second group, Multiple System Operators (MSOs) own headends and infrastructure but are not as distributed as Telcos. Additionally, these parties have lower sunken costs and face fewer regulatory restrictions, creating a lower cost base than the Telcos.

Lastly there are the hyperscale operators with vast, efficient data infrastructure and technical expertise available and investment capacity. However direct access is offset by a weaker local infrastructure position and deployment takes time. In the USA this disparity has resulted in marriages of convenience where one party has an infrastructural footprint and another has the technology to make connections.

It's important to realise that edge isn't a magic solution for every application. Finding the right balance between central and local is essential. For services such as Netflix, it makes sense to centralise the bulk of files to avoid huge numbers of storage devices at every location, whilst using the edge for streaming so that users have a seamless experience. Another example exists with autonomous vehicles: it makes sense to store maps centrally and download as required, whereas the data and processing required for split-second decisions in potentially dangerous situations remains as close to the vehicle as possible.

What's more, we need to realise that edge data centres come in many shapes and sizes ranging from fridge-sized units at the base of cell towers supporting 5G, to central office-sized configurations, all depending on the application. Box based edge DCs will have just a few panels and connectors, whereas the larger installations will require the same product decisions as other enterprise DCs needing to connect rows of cabinets challenging the economics of copper and fibre.

In the near future there will be many more immense hyperscale facilities that coexist with millions of small edge DCs. These may well be segmented into standardised small, medium and large classifications. For the passive network industry aiming to provide optimised cabling solutions, it is important that some standardisation exists and that rules are now set.

Finally, we should remember that although edge is often linked to 5G, which is often considered to be the key driver, once fully-fledged edge networks are in place, completely new applications will arrive which we probably can't even imagine today. As a result, current infrastructure decisions need to be as open-ended as possible. NC