The debate over the need for computing at the edge has been centered around low latency for advanced applications, automation and real-time decision making.
Our collective thinking has led us to believe that if we enable cloud-native capability close to the source of data generation, shortening the distance the data must travel, applications can process them faster, deliver decisions quickly and deliver a superior user experience. Viewing edge use cases through a narrow lens of latency requirements has led to the debate over the need for micro data centers near cell towers, with the presumption that they are the closest to the devices delivering the lowest latency.
However, a more fundamental shift is happening that is driving the need for edge computing beyond low latency. While reducing latency is important, there are lot of other important drivers at play for infrastructure decisions.
From Consumption to Co-Generation
In the last decade, we solved the problem of consumption with high-capacity networks that made it possible for us to binge on HD videos during the Covid lockdown period. The challenge was that of dissemination, where content creators distributed content from one source to many consumers. Big fat pipes of network capacity with some amount of caching enabled moving data to many consumers at faster speeds.
In this decade, we have a fundamentally different problem that is emerging. Billions of devices connected to the network will generate data 24/7, the problem to solve is now one of aggregation. Disparate and distributed data needs to be aggregated and processed to derive intelligence. While delivering that intelligence in a few milliseconds remains a problem, managing that data is the more significant problem. The logistics of transporting, processing, and securing the data present a huge operational challenge.
The Data Logistics Problem
In the physical world, logistical problems are addressed by having full control of the operational environment. Having compute and storage resources within a reduced perimeter and having operational control over the data perimeter helps mitigate much of this logistical challenge.
I submit the following guiding principles for locating edge computing resources.
1. Understanding the data.
Knowing the sources of the data, the rate at which it is generated and/or consumed, its shelf life, security, governance, and statutory requirements.
2. Control of the operational environment
Likewise, it is important to evaluate the environment in which the data is going to be moved as well as the requirements to have full operational control of that environment. Network transport, the data center that houses the compute resources, the cloud services, the application, and the stakeholders will need to evaluate to achieve desired control by way of contracts and SLAs.
3. Operational Cost
The cost of compute resources in a hyperscale cloud is always going to be cheaper than at the edge. But the operational cost of moving the data within a secured network to a faraway cloud could be more cost-efficient at the edge. With consistent edge availability, the savings will soon outweigh the operational cost of managing the data logistics.
Location of compute resources should be looked at as a function of operational control while at the same time delivering the latency and bandwidth requirements of the application. By way of example consider a service provider, delivering security as a service with advanced video analytics using AI. While it is technically feasible to deploy the application in the cloud and deliver intelligent alerts to a user in seconds, the service provider will have significant operational challenges. To implement operational processes for transport, security, archival and governance with multiple stakeholders to comply with SLAs, thousands of cameras streaming data 24/7 from multiple customers, operations will be complex and expensive. But processing the data within the confines of the service provider network gives them full operational control they need to manage the services.
In summary, we must expand our view of why edge computing is needed. By looking at the edge as a means for solving the complex problem of aggregation of disparate, non-stop and time-sensitive data and inferring intelligence. Investing resources in this exercise and understanding the logistics of managing the data is important. When done right, applications will find their edge. The desired latency will be a natural outcome.
About the author
Venky Swaminathan is Founder and CTO of Trilogy Networks. Learn more about Trilogy Networks and how we can help with your data logistics problems at Follow us on Twitter and LinkedIn.