Deploy AI Applications to a Fleet of Nodes: A Use Case

​

Introduction

Deploying an AI workload to dozens or hundreds of nodes is a common operational challenge. This article walks through the approaches available in Edge Infrastructure Services for deploying AI application instances at scale, from targeted single-node deployments to fully automated fleet-wide rollouts using the App Profiles API.

​

Deployment Approaches

​

Deploy to individual nodes

The standard approach for smaller fleets or when each node requires a distinct configuration is to deploy one application instance per node through the console.

1. Navigate to Infrastructure > Workloads > Applications.
2. Select the contextual + create action.
3. In the wizard, select the edge app image, select Node as the target type, and pick the target node.
4. Complete the remaining steps (interfaces, custom configuration, features) for that node.
5. Repeat for each node.

This approach gives you full control over per-node configuration and interface assignments, and it is the right choice when node-to-node configuration differs significantly (for example, when each node connects to a different local network instance or uses different custom configuration values for a site-specific parameter).

​

Deploy to a cluster

If your fleet uses clusters managed by the ZEDEDA Edge Kubernetes Service (ZEKS), you can deploy a single application instance to the cluster and let the cluster scheduler handle placement across member nodes.

1. Navigate to Infrastructure > Workloads > Applications.
2. Select the contextual + create action.
3. Select Node Cluster as the target type and pick the target cluster.
4. Optionally pin the instance to a preferred node within the cluster and configure failover behavior.

Cluster deployments reduce the number of instances you need to manage directly and provide automatic failover if a member node goes offline. For a full walkthrough of the deployment wizard, see Deploy an Application Instance.

​

Automate fleet-wide rollouts with App Profiles

For large-scale, repeatable deployments where many nodes should receive the same application configuration, use the App Profiles API. App Profiles are not visible in the Edge Infrastructure Services console, but they are accessible through the REST API and the Terraform Provider for ZEDEDA Cloud. An App Profiles deployment uses three objects:

• App Profile: defines the application, its network configuration, and volume settings.
• Asset Group: a logical grouping of the nodes that should receive the deployment.
• Deployment Profile: associates an App Profile version with an Asset Group and triggers the rollout.

When you update a Deployment Profile, Edge Infrastructure Services rolls out the change to every node in the Asset Group automatically, without requiring a per-node trigger. This makes it well suited for AI model updates, configuration changes, or version upgrades that need to reach the entire fleet consistently. For a complete walkthrough with API and Terraform examples, see Automate Deployments with Application Profiles.

​

Recommended Pattern for AI Fleet Deployments

The following pattern works well for deploying and updating AI inference workloads across a large fleet:

1. Create the AI app image in Infrastructure > Workloads > Build > App Images. Define custom configuration variables for any parameters that change per deployment (for example, a model endpoint URL or inference threshold).
2. Group nodes into an Asset Group by project and tag (for example, region=west-coast, type=inference-node). This lets you target a subset of your fleet without managing individual node IDs.
3. Define an App Profile that bundles the AI application, its network policy (for example, a local virtual network for model data traffic), and any volume policy for model storage.
4. Deploy the Deployment Profile to the Asset Group. Edge Infrastructure Services creates the application instance on each member node and reports status per instance.
5. Update the App Profile when a new model version is ready. Incrementing the profile version triggers an automatic update to all nodes in the Asset Group. You can optionally configure the profile to take snapshots before updating so each node has a rollback point.
6. Monitor the fleet from Infrastructure > Workloads > Applications. Filter by application name or node to track which instances are online, updating, or in an error state.

​

Considerations for AI Workloads

• Image size: AI models can be large. Ensure each target node has sufficient storage before deploying. Check available storage from the node detail page under Storage.
• CPU placement: For latency-sensitive inference workloads, set the edge app’s CPU placement policy to the Dedicated or Isolated pool. This reserves physical cores for the application and reduces scheduling interference from other workloads. You can override the policy per instance when you deploy. See CPU Placement for Application Instances.
• Network isolation: Deploy a local virtual network on each target node to isolate inference traffic from management traffic. See Create a Virtual Network for setup instructions.
• Snapshot before updates: Take a snapshot before each model update so you can roll back to the previous version if the new model causes regressions. See Application Snapshot Overview.

​

Next Steps