Are there definitions of the different edges for Industrial ?
And how many edges are there ?
IMHO, the "edge" isn't really an edge. The cleanest definition is "anything outside a data center", aka the entire real world. But that's not really helpful.
I see the edge as more of a "slope", where the height is determined by (connection) proximity to the essentially infinite compute and storage resources of the cloud. Those resources make high-end computing like AI easier. Some assume that edge systems are less intelligent. But practically, many of the most useful intelligent systems must implement intelligence further down the slope. Autonomous vehicles are a great example. They must do sensor fusion, localization, planning and control directly inside the vehicle.
"Edge is in the eye of beholders!" If you are a qualified system architecture, you should know clearly which parts are considered "edge"; otherwise, you may need to look for a real job.
Most often, different trade has formed, through their common practices, a common "understanding" about where an "edge" is. For example, I have seen the "Data Center" community indicate they are "the edge" in the Cellular Network Comminity in the most recent "Edge Congress" meeting I chaired in Austin. The view is already different than Stan's thought already.
This is partly a sneak preview of a technical report we are working. Referring to the "outside the data center" definition for edge computing (note, I'm not using the term edge), a key differentiator is how the compute and network infrastructure are managed. Data center compute infrastructure is very uniform and secured within a physical location, whereas edge compute infrastructure is usually greatly scattered, highly heterogeneous and very challenging to secure, upgrade and monitor. Mitch points out above that many vendors are looking to provide cloud compute assets smaller in size and located at the end of high bandwidth networks. From a management perspective these are closer to data centers than edge devices. This is indeed a gray area, but I think the comparison with data centers helps to define edge computing at least.
>Are there definitions of the different edges for Industrial ?
From English, edge is a transition point; e.g., the edge of a table or the edge of the ocean. In IoT there are several important edges, including:
1- The sensor, where the world meets cyber space (e.g., temperature is sensed into a digital value).
2- Where the direct link or the LAN meets the WAN, usually in the IoT gateway. The security posture on one side is very different from the other, thus an important transition point.
3- Where the cellular wireless meets the fiber backbone at the cellular tower; what the Telcos call MEC for Multi-access Edge Computing.
In general, a tree hierarchy of compute aggregation tiers (aka, multi-edges) that goes from the numerous to the few could be an interesting organizing paradigm for certain use cases.
In addition, now cloud computing is equally applicable at the edge as it is in the data center; for example check "edge cloud" or this week's Amazon's announcement of "Wavelength" and "Outposts". As was shared already, the interesting dichotomy is "data center" vs. outside of one, we call edge computing ("Field" would have been a better term, but the world already chose "Edge"). Cloud computing could be in either and really means we do not know nor care where. The report that Brett alluded to and the Distributed Computing TG is working on gets into the details.
The economies of volume suggest that it is cheaper (both CapEx and OpEx) to do computing in the data center; however, in certain situations (IoT being one of them) this saving is not feasible. We narrowed down the reasons to the following:
1- Issues with the link to the data center:
a. is unreliable (dropped connection),
b. has too high a latency for a given application,
c. has insufficient throughput for a given application, or
d. is too expensive for the given data volume.
2- Data residency, governed by policy and/or regulation; e.g.:
a. Data from sensors not to leave the premises (trade secret, HIPAA privacy, ...).
b. A user interface is restricted to a premise (e.g., SCADA systems).
In those cases, as Brett said, you have to bite the bullet and locate computing outside the data center, where it will be more costly and has special requirements including hardening , tamper resistance, and it will be more difficult to manage and maintain. Effectively, the location of the compute resource becomes part of the problem requirement and solution design.
To me what telcos call MEC, compute resource at the cellular tower, (where the fiber meets the wireless) is a form of public edge computing (in contrast to the private variety). For one they refer to it as "edge", the "E" in MEC, and it fits the reasons above else it would be more economical to locate it in a more centralized data center, instead of at every cellular tower location in the wild field exposed to the elements.
Consult the recent definitions of "edge" and "edge computing" in standards, e.g. ISO/IEC TR 30164 INTERNET OF THINGS (IoT) – EDGE COMPUTING