Kubernetes has become a popular platform for running machine learning workloads. Typically, discussions about adopting Kubernetes start with questions about how well the cloud provider supports it. But more and more, we are seeing Kubernetes being run on the edge — that is, on devices not in large data centers, but in various settings such as a retail stores logging sales volumes, a manufacturing floor collecting metrics from manufacturing systems, in hospitals processing patient data, or a warehouse helping keep track of inventory. It is important to point out that edge devices will not replace Kubernetes or microservices in the cloud, but there are certain workloads that can benefit from running Kubernetes on the edge.
Kubernetes is a container orchestration platform that enhances the efficiency of application development. It ensures robust application deployments by restarting them in case of failure, provides high availability, and facilitates easy application updates with the ability to roll back if needed.
Recently, Brian Chambers, Chief Architect at Chick-fil-A, described how they deployed Kubernetes to all 3000 of their restaurants. Kubernetes has enabled him and his team to develop and deploy applications ranging from collecting telemetry from smart cooking appliances to tracking point of sale transactions. They have been able to use this aggregated data to ensure the quality of the food that is being prepared in their restaurants, measure sales performance and the speed of their service, and generate customer volume forecasts among other things. Due to the limited availability of highly available internet connections in most fast-food restaurants, sending all this data to the cloud for reliable aggregation and processing would not have been feasible.
Another example of this is vision AI at the edge, which involves using artificial intelligence to detect objects in real-time video feeds. While uploading a raw video stream from a few cameras to a cloud provider may be simple, it becomes a cost challenge when dealing with 100 or 1000 cameras. Having a device on the edge process the video and only upload points of interest for further processing greatly reduces the amount of data that has to be ingested by the cloud platform. This, in turn, reduces the required bandwidth and lowers the potential cost of transmitting that data to the cloud. Increasingly, we see edge and IoT devices generating data that can’t be reliably transported to the cloud, but this data can still be utilized with edge computing.
Even though there are many benefits, running Kubernetes on the edge does come with some unique challenges. One of them is that cloud providers typically manage certain Kubernetes components that would have to be manually deployed on an edge device. This is a minor hurdle since there are multiple production-ready projects available to fill these gaps. Some of these same projects are even utilized by the large cloud providers, just modified and re-branded to fit their platform. Despite this, it does add a bit more complexity to the already complex solution that Kubernetes can be.
The versatility of Kubernetes extends beyond traditional cloud environments, finding a pivotal role on the edge. The practicality of running Kubernetes on the edge becomes evident in scenarios where reliable internet connections are scarce, or when large amounts are data are too costly to transport. Moreover, the increasing prevalence of edge and IoT devices underscores the importance of local data processing, optimizing bandwidth usage and minimizing costs. Kubernetes on the edge emerges not as a replacement but as a strategic augmentation for specific workloads, enhancing data efficiency and operational agility.
About the Author: "Angel Garza is a Cloud Enthusiast | Technology Advocate with a lot of passion to add value to the R&D and AI ecosystem. Angel is developing his expertise in Kubernetes, Networking and Virtual Machines and our Technical Leadership has been very impressed with his work. On behalf of Katalyst Street, we would like to thank Angel Garza for this guest blog and look forward for more Katalyst Street guided contributions in the future."