Verizon, Google highlight how 5G is changing the cloud

Nvidia GTC21 5G AI edge panel
Shailesh Shukla, VP and GM of Networking and Telecom at Google, said knowing where exactly the edge is and what workloads to put there can be tricky. (Screenshot, Nvidia GTC21)

Executives from Verizon, Google, Mavenir and Wind River stressed the importance of artificial intelligence (AI) in enabling a 5G world and spotlighted the ways in which the need for deployments of the technology at the edge was transforming the cloud landscape.

During a panel session at Nvidia’s GTC21 conference, Verizon Media’s VP and CTO of 5G MEC, AI Platforms and Next-Gen Applications Ganesh Harinath said AI is “not a want, it’s a need” as operators pursue next generation applications.

While the industry has been running AI and machine learning in the cloud for close to a decade, Harinath explained the landscape is now shifting to meet the needs of a new class of applications. “The undercurrent is the inferencing aspect of machine learning has to be moved closer and closer to where the signals are generated” by devices like video cameras and sensors, he said.

Mavenir CEO Pardeep Kohli noted sensors are generating increasing amounts of data. To process massive amounts of information at a central location would require more and costly transport, so “moving it closer to the edge where you can not only capture the data and process it and using AI tools actually act on it” opens the door to new possibilities as far as applications go, he said.

But Shailesh Shukla, VP and GM of Networking and Telecom at Google, said knowing where exactly the edge is and what workloads to put there can be tricky.

RELATED: Google Cloud, AWS and Azure want to 'friend' telcos, but not every telco is ready to click 'accept'

He noted the word “edge” is an amorphous term which can refer to anything from the IoT edge, enterprise edge, telco edge (points of presence, or POPs) and the public cloud. The decision about which of these applies will depend on the application and corresponding considerations around latency, cost of backhaul and other requirements.

For a use case focused on identifying product defects in manufacturing, for example, Shukla said it wouldn’t make sense to train the necessary algorithms on the factory floor. That would instead be done in a centralized cloud capable of offering a massive amount of compute and storage. The actual application of that algorithm for inferencing and identification, however, would run on the edge.

With this in mind, Wind River head of Telco Solutions Nermin Mohamed said “in order to make 5G really successful, we need a highly disruptive cloud model where real-time compute” is pushed to the “intelligent edge.”

RELATED: Words of wisdom from edge experts

Shukla argued in favor of a platform approach, noting a common control plane spanning all spokes of the edge would make it possible to “place the right kind of workload at the right place” to deliver any use case. “If you had completely different control planes or completely different applications that you had to build for each location, suddenly it becomes a lot more complicated. So our view is that working with the telecom operators and the broader ecosystem you can create applications once, deploy them anywhere and run the right kind of workload at the right place.”