Canonical’s Sessions at GTC
Simplifying Kubernetes across the Clouds: MicroK8s on NVIDIA Tech Stack [SS33138]
Although Kubernetes revolutionised the software life cycle, its steep learning curve still discourages many users from adopting it. MicroK8s is a production-grade, low-touch Kubernetes that abstracts the complexity and can address use cases from workstations to clouds to the edge. We’ll highlight the details of MicroK8s’ simplicity and robustness and demonstrate the different usage scenarios, running it on NVIDIA DGX, EGX, DPU and Jetson hardware using real applications from NVIDIA marketplace.
Speaker: Alex Chalkias, Product Manager
The Power of Micro Clouds, a New Class of Compute for the Edge [SS33238]
Is edge computing reversing years of cloud investment? What constitutes a strong edge strategy?
Iteration of cloud technologies has made automation simpler than ever and created the right primitives to solve all kinds of problems. The reality, however, is more distributed than cloud computing is. Devices can be anywhere, in motion, or even sent in someplace to never come back. These challenges apply to edge data centers (with NVIDIA EGX boxes) as well as smart devices (with NVIDIA Jetsons) with increasing GPU usage for low latency analytics, inferencing, and visualization, while offloading security and storage to DPUs.
With micro clouds, it becomes possible to bring cloud capabilities to the broadest range of devices at the edge, accelerating innovation and underpinning operations.
This talk will cover Canonical’s micro cloud ingredients for the edge, an opinionated open source stack, from Ubuntu to MicroK8s, that gives developers the same virtualization and containerization capabilities from cloud to edge.
Speaker: Valentin Viennot, Product Manager
GPU Workloads on a Mixed Container and VM Infrastructure [S31723]
We’ll cover running a small-scale (three to 50 servers) infrastructure including GPUs and making those available to various users through either system containers or virtual machines. LXD is a simple yet powerful piece of software allowing deployments like this — it supports easy clustering, multiple independent projects, and can run both container and virtual machines, mixing them on every one of the nodes. This, combined with its support for GPU pass-through to both containers and virtual machines, makes it possible to get the most out of your physical resources. We’ll go over what LXD is, how clustering and projects work, and what the supported options for using GPUs are, including a demonstration of such a deployment.
Stephan Graber, Project leader for LXC/LXD Christian Brauner, Kernel Engineer