Skip to content

Mokka on DSX Air

Use the hosted Kubernetes Intro – Mokka + KWOK demos on DSX Air to explore the Kubernetes GPU stack without physical GPUs or a local repository checkout. Choose a configured environment to explore the stack, or an unconfigured environment to work through its setup.

What DSX Air provides

NVIDIA DSX Air is a cloud-hosted data center simulation platform for validating infrastructure configurations, features, and automation. It lets you build a digital twin of an infrastructure environment and experiment with network topologies and operating systems.

Its Demo Marketplace provides ready-made simulations with accompanying instructions. Launching a demo creates a copy in your account that you can modify and rebuild. The Mokka + KWOK demos use this delivery model to provide a hosted environment for exploring the Kubernetes GPU stack.

How the integration fits together

The demos bring together three layers:

Layer Role
DSX Air Hosts the demo environment and its lab instructions.
Mokka Supplies simulated GPU driver interfaces on CPU nodes that run containers, so real tools such as nvidia-smi and GPU management components can discover simulated devices. See the Mokka architecture.
Kubernetes WithOut Kubelet (KWOK) Represents virtual Kubernetes nodes without running a kubelet or workload containers on each one, allowing control-plane exercises with a larger node population.

These layers exercise different parts of the stack. A KWOK node's Kubernetes status does not demonstrate container execution. Use a Mokka-backed node that runs containers when inspecting driver interfaces with nvidia-smi.

Access DSX Air

DSX Air uses NVIDIA NGC for authentication and organization management. You need an NGC account and organization; sign in with your business email and select the organization you want to use. Each user needs an organization-level Air User or Air Org Admin role. See Account Setup for onboarding and access troubleshooting.

Choose a demo

Demo Use it to
Kubernetes Intro – Mokka + KWOK Configured Explore the configured environment.
Kubernetes Intro – Mokka + KWOK Unconfigured Work through configuration using the demo's lab instructions.
  1. Open the selected demo and click Launch to create your copy. Deployment can take several minutes; find the copy in your Simulations list.
  2. Open the demo's Documentation for credentials, cluster access, component versions, and setup steps. See Pre-built Demos for the launch workflow.
  3. In the simulation's Topology tab, wait until its status is Active, then double-click a node to open its console. Select the node identified by the lab instructions for Kubernetes access. The DSX Air Quick Start also explains node properties and external connections through the Services tab.

For the configured demo, inspect the existing deployment before making changes. For the unconfigured demo, complete the lab's setup before checking for simulated GPUs. The installation reference explains Mokka's deployment options if you need more context for those steps.

Inspect the environment

Run these commands in the lab terminal with kubectl configured for the demo cluster:

kubectl config current-context
kubectl get nodes -o wide
kubectl get pods -A -o wide
kubectl get nodes \
  -o custom-columns='NODE:.metadata.name,GPUS:.status.allocatable.nvidia\.com/gpu'

The node and pod listings help identify where components run. The last command shows advertised nvidia.com/gpu resources, if present. That resource count is an allocation view; it does not establish whether a node runs containers or provides Mokka's driver interfaces.

After Mokka is configured, find one of its node-daemon pods in the pod listing. Replace <namespace> and <mokka-pod> below with its namespace and pod name:

kubectl -n <namespace> exec <mokka-pod> -c node-agent -- nvidia-smi -L

This checks the driver view on that pod's node. The reported devices come from the active GPU configuration; use the demo's instructions for its expected model and count. If the command fails, consult Troubleshooting.

For the scope of what these checks prove, see Mokka's simulation boundaries. Use the device-plugin guide for GPU resource allocation and runtime control for changing simulated device state.

Save and resume the lab

When finished, select Stop Simulation and Store Checkpoint to save the nodes' state and release compute resources. Starting the simulation again resumes from its latest checkpoint by default. This lets you continue a setup exercise later. See Simulation Management for checkpoint selection, automatic sleep, and lifecycle details.