Advertising IoT Devices as Extended Resources in Kubernetes w/ Akri

Kubernetes is powerfully declarative. When deploying a Pod, you can specify memory and CPU resource limits, so your application will have all the resources it needs.

A few years ago, Kubernetes created the Device Plugin framework to allow other resources to be requested. It was created with the cloud in mind, particularly focused on allowing static hardware resources, such as GPUs, to be requested by Pods.

On the edge, Kubernetes workloads are utilizing a variety of IoT devices with intermittent connectivity.

🔹 What if these devices could be requested as resources just like GPUs or CPUs?

🔹 What if IP devices could be connected to the network and immediately seen as a requestable resource?

🔹 What if workloads could automatically be deployed to discovered devices?

Akri (A Kubernetes Resource Interface) is an Open Source project that brings the Device Plugin framework to the edge, enabling these IoT devices to be requested as resources and automatically deploying user-provided workloads that utilize these devices.

This talk discusses the power of Kubernetes’ declarative approach to requesting resources and how Kubernetes concepts that were created for static cloud environments can be leveraged on the edge where the environment is dynamic and ever-changing.

Agenda:

🔺 5:00 – 5:10 Introductions

🔺 5:10 – 5:40 Presentation

🔺 5:40 – 6:00 Q&A

We’ll cover:

🔹 Kubernetes Resources

🔹 Traditional use of Device Plugin framework

🔹 Unique Edge/ IoT environments

🔹 Exposing IoT devices as requestable resources using Akri

🔹 Automating and simplifying usage of discovered devices with Akri

🔹 Demo of using Akri to discover mock cameras

Speakers:

🔶 Kate Goldenring, Software Engineer (Microsoft)

🔶 Vijji Suryadevara, CPO & Head of Engineering (Phizzle)

Event Details:

ZOOM WEBINAR LINK:

https://zoom.us/webinar/register/WN_4uJS1nseTRCg1Hzi9raqnA

Date & Time:

Thursday, April 29, 2021

5:30 AM to 6:30 AM IST

PrevPreviousGitOps and Progressive Delivery with Flux, Istio, and Flagger

NextEvents first: cloud-native patterns beyond request/responseNext