Don’t go overboard at the edge in edge computing

In many cases, centralized control in the cloud makes more sense than trying to do everything at the edge

The whole idea of edge computing is to push some processing and data retention out to the edge of the network. As we move to internet of things (IoT) devices that are gathering gobs of data, it makes sense to process that data close to where it’s gathered. In some cases—like managing a jet engine—you want to return results directly from the edge device, reducing the need for bandwidth and decreasing response latency.

Today, most major public cloud providers are pushing their own edge computing services. But pushing everything out the edge is not always possible—and not always a good idea.

If the cloud provider is going to manage edge computing using a set of cloud services, and do so in a scalable way, there really needs to be central control. This includes security, governance, and management of those edge processors. While this central control seems to fly in the face of what edge computing is as a concept, there is really a good reason for these having such control.

Although edge computing means that there are independent processing existing at the edge, the reality is that moving processing to the edge is not always a good idea. Case in point is deep analytics services that are processor- and I/O-intensive; it can’t happen at speed on the thermostat on the wall, in the robot on the factory floor, or even in your own automobile. So, the data is centralized, as is the processing, and results return to the edge. In such cases, it really should not matter where the processing occurs.

Also, keep in mind that you need to maintain virtualization of these processes, whether at the edge or centrally. So again, it should not matter where things are processed, as long as there is technology keeping track of what’s where and what’s doing what. It makes more sense for that tracking technology to be in a public cloud platform that is centralized by nature.

The power of edge computing is really not where the data and the processing reside, but that the processing can be separated if needed and managed centrally. I don’t get religious around where things should run; I don’t think you should either.