It is too early to say how edge computing might grow as a percentage of overall computing cost. or cloud computing cost. Centralized cloud computing has pushed more of the workload cost to remote data centers and away from on-device or local area networks (both examples of edge computing).
But, historically, computing cost has arguably been concentrated at the edge, on board devices or on the premises. The shift to cloud computing has moved workload locations to “far edge,” so perhaps the better analogy is local versus remote computing.
It is not yet clear how much the balance of capital investment will change over time, though edge investments will grow. The issue is not so much edge computing “on the premises” or “in the cloud” but likely how much of the workload moves to some “close to the premises” location.
The reason is simple. Hybrid computing exists because many enterprises use a mix of owned infrastructure on the premises and cloud computing. What is new is a migration of some of that investment to new “edge” deployments, which might be on premises, near the premises or at other locations in a metro area.
With one exception, enterprise workloads would seem to be key, as most “north-south” traffic from data centers to end users consists of video. Live videoconferencing will not benefit as much from edge computing, which remains a multipoint-to-multipoint, episodic use case.
Entertainment video often benefits from edge caching, which arguably is a form of edge computing.
The trade offs might be similar to communications network investment. Communication network costs always are disproportionately concentrated on the network edges. The past rule of thumb was that the access network represents about 80 percent of total capital investment and perhaps a higher percentage of operating cost as well.
Among the reasons is resource sharing, which is highest in the core, least at the edge. Generally speaking, the greater the degree of sharing, the lower the per-customer share of cost. What differs is the percentage of cost incurred by service providers compared to end users.
Most enterprise data generation happens at the edge. The issue is data processing.
If it is correct that, eventually, most enterprise computing happens at the edge, then one should also expect computing investment to shift to the edge. Whether that is “back to the traditional edge” (enterprise premises) or “to some new edge location” (off premises, but within a few miles) is what will be determined.
No comments:
Post a Comment