Edge Value and Cost: Lots of Moving Parts

The oft-stated value of edge computing is support for ultra-low latency applications. Still, as content delivery networks have shown, reduction of wide area network load or cost also can be a primary value of edge storage or computing. 

But improved isolation and security, possibly to support regulatory or corporate compliance rules, also provide value. Workloads running on-premises do not send data into the public cloud. Edge-processed data also can be obscured, transformed, or encrypted prior to sending it upstream to a far-edge data center for archiving or backup. 

Improved jitter performance is another advantage of edge computing, for applications that require predictable packet arrival. That might be similar to the value provided by edge caches of entertainment video, which do not traverse the WAN, and which generally consist of non-real-time content. 

Edge facilities might not help for video conferencing using the public internet, which must, almost by definition, traverse the WAN. 

In some instances, where connectivity suffers from intermittent availability, on-premises processing provides continuity. Venues such as cruise ships, airplanes, oil rigs and vehicles provide examples. 

But those advantages come at a cost of higher per-cycle or per-workload unit costs. Far-end hyperscale data centers benefit from huge economies of scale. Retail pricing for Amazon Web Services “Wavelength,” which supplies AWS functionality at the edge is about 20 percent to 30 percent higher than the same operations conducted at an AWS region facility, for example. 

Analysts at AvidThink note that the cost of an on-premises AWS Outposts deployment--which drops a rack of AWS servers into an on-premises data center, runs between 20 percent to 50 percent higher than the equivalent volume of workloads conducted remotely at an AWS hyperscale site. 

So the decision to use remote far end or on-premises edge computing is not simple. Information technologists have to consider:

• Where the data is generated

• Where data should be processed

• Where processed data needs to be consumed (by end-users or other services)

• Where information is eventually stored

• Performance needs of the application in terms of response times, throughput

• Security and compliance

• Capabilities of the underlying infrastructure platform, especially if specialized hardware (GPUs or FPGAs) are required

• Cost of data transport

• Cost of transient and permanent storage of data

• Cost of computing and memory at each location

• Availability constraints for the application and reliability of infrastructure at each location

• Budget issues

The value and cost of edge computing therefore is not a simple matter of functionality or cost.