Edge computing is coming for any number of reasons, including some we do not necessarily and routinely tout, such as support for compute-intensive operations such as artificial reality and other forms of extended reality; real-time process control or any applications that require ultra-low latency.
Consider power efficiency and footprint.
“Even with 5G right now, which is a little more power-efficient and has a lot more programmability of the network, from a sustainability point of view, it is impossible to continue transmitting everything to the cloud,” says Stacey Shulman, Intel VP, network and edge computing.
The actual benefit remains to be verified. Some argue that hyperscale computing facilities and instances are inherently more efficient than small distributed instances. But that only accounts for power consumption per million cycles, and not the full footprint.
Also, some use cases might require local computing, and the footprint simply must be optimized. The cost of a computing instance or its environmental impact is arguably always secondary to the value of the function: keeping cars from crashing and injuring people, for example, when operating autonomously.
And different processing tasks consumer different amounts of power. Still, it has been argued that power consumption does not scale linearly with the volume of computing instances.
If one considers applications that require the use of artificial intelligence in real time, “here’s a lot of data that can’t go straight to the cloud,” she notes.
Some analyses already suggest that bandwidth cost savings are a benefit of edge processing. Power footprint might well wind up as among the key benefits of edge computing, Shulman argues.
Some argue edge computing should be more efficient. But the total impact has to include operating the communications infrastructure, adjust for the types of workloads and utilization of server resources, for example.
In the past, precisely the opposite argument has been made. AWS has argued that
remote cloud computing is more efficient, and has a smaller footprint, than “premises” computing at the edge.
As always with identifying the precise impact of any activity or product, the analysis is complicated and requires many assumptions.
So far, modeling has focused on computing footprint at hyperscale data centers. We do not have enough data, yet, on edge computing instances, except to the extent that we can assume minimum and maximum power consumption for a single physical server or virtualized workloads run on a shared server.
The full footprint then will hinge on other inputs. Some might also include any substitution effects, where using edge computing reduces the footprint of other activities that are displaced or lessened.
The point is that we still are not sure how edge computing will wind up, when compared to remote cloud computing, in terms of carbon and other greenhouse gas footprint.
No comments:
Post a Comment