Nodes

A view of Nodes connected to the system is presented here, as well as the ability to add new nodes, either with direct connection or by using the AiQuAgent docker image on the node.

node-list

The node list shows each node’s Status (online/offline indicator), ID, Name, Host address and GPU count. If your realm has access to shared nodes, they are listed under Global Nodes (which you can only enable or disable for your tenant), separate from your own Tenant Nodes.

If you click on a (tenant) Node in the table you can:

  • Enable or disable individual GPUs on the node.

  • Change Node Information:

    • Public address the node is reachable on

    • Docker port on the node

    • Host — the internal address where the node is reachable by AiQu

    • Name of node (shown in the rest of the AiQu system)

  • Select the node’s Node Type(s).

  • Restrict traffic with Local Blocked Networks and Local Allowed Networks (CIDR notation).

  • Manage System Jobs on the node with Restart and Restart & Update actions.

  • Edit the node’s Data Collection Config.

If you want to add a new node you have two options:

  • Add node: Add a node that is already able to communicate directly with AiQu via a form. This usually requires a VPN connection.

  • Connect new node (recommended): Generate an invite and a docker command that you run on the node to establish a connection to AiQu. This approach makes it less likely for the connection to be blocked by the firewall, since the connection is initialized from within the network.

Create Node Invite

create-node-invite

  • Label: A label identifying the invite/node

  • GPU’s per node: Number of GPU’s per node

  • Valid until: The date after which the invite can no longer be used

  • Max uses: How many times the node invite can be used

  • GPU link type: How GPUs are linked on the node (e.g. all connected, dgx1 or custom)

After creating an invite it appears in the invites list with its ID, Label, Valid until, GPU link type, Max uses and GPU count. Run the generated docker command on the target node to connect it.

* For the nodes to be able to run jobs they have to have docker installed. Some jobs may require other software to be installed. For example: to run images pulled from NGC on Nvidia GPU’s the Nvidia Container Runtime.