“Where is your edge?” always is a starting point whenever edge computing investments and strategy are concerned. For content delivery networks, “edge” was “someplace in the metro area.” For manufacturing plant internet of things, the edge computing might happen on the premises.
For some devices, the edge is the device itself, as processing is onboard. That might be true for autonomous vehicles or any number of consumer appliances.
For would-be suppliers of infrastructure edge computing (edge computing centers operated or owned by connectivity suppliers or third parties), the range of physical decisions is greater, and there are practical issues as well.
How to pace investments with expected demand always is an issue for infrastructure-based businesses. And those sorts of business issues might well dictate “where the edge is,” in the early days of infrastructure edge computing.
For any communications network, most of the cost is close to the edge, in the access network and devices used at the edge. Capital intensity declines as one moves back into the core of the network. That will be true for edge computing as well: most of the cost will be at the edge.
Less investment will be needed if the number of computing sites is reduced by a factor ranging from three to 10 or even 100. That is the case if edge computing initially happens at aggregation sites instead of at cell towers, for example.
And many of the issues common for small cell placement also are issues for edge computing sites: the availability of a fair amount of optical fiber backhaul. Any edge site expecting robust traffic growth will require surplus optical fiber backhaul capability. In other words, lots of available fiber.
Until recently, few mobile or fixed network operators in the access business have deployed lots of excess fiber for future growth, though metro capacity suppliers might do so routinely. Still, as a rule, companies carefully match capital investments with the expected use cases and revenue opportunities.
And it is fair to say planners did not anticipate the needs either of ubiquitous small cell networks or edge computing centers.
Also, third party edge computing suppliers will likely have to have interconnection agreements with underlying transport providers, so that signaling and traffic are kept local, instead of having to first move deeper into the network. In other words, the whole point of edge computing is to process traffic locally, rather than across the internet to remote data centers.
Even at the regional level, it will not always be helpful if signaling and bearer loads have to move from the access edge to a regional center first, and then be sent back to the access edge for processing, for example.
And, as always, there are the capital availability issues. At least initially, there might not be enough demand, or enough available capital, for any single entity to deploy an edge computing facility at every cell site, for example. It might make more sense to match projected demand with investment by deploying a thinner network at central offices or other aggregation points, instead.
As with the general pattern for cell sites, densification happens over time as demand grows. Operators do not densify first, they do it later, as volume grows. Right now, enterprise demand for infrastructure edge computing is not so high as to warrant “over-investment” in edge computing facilities.
Again, “where does the edge computing need to be?” is a relevant question. As has been true for connectivity providers in the analog and digital eras, on fixed and mobile networks, switching occurs as a “tree” or hierarchical architecture.
That also seems to be the case for cloud computing, with the remote hyperscale data centers complemented by more regional centers, then eventually edge aggregation sites and then access edge facilities.
