Skip to main content
AI Data Engine

Requirements for cabling data compute nodes for AI Data Engine

Contributors netapp-jsnyder netapp-dbagwell netapp-driley
PDFs

Data compute nodes integrate with your AFX 1K storage system through host network and cluster network connections. Review the I/O slot configuration, cable types, and connection requirements for your deployment.

Cabling configuration

Data compute nodes connect to the same cluster switches as the AFX 1K controller nodes, extending your storage system with compute resources optimized for AI and machine learning workloads.

The initial AI Data Engine (AIDE) configuration supports a minimum of three data compute nodes. For comprehensive configuration details and slot priorities, see NetApp Hardware Universe.

Slot numbering on a data compute node
Callout number 1

Unused slot on the data compute node.
Callout number 2

Unused slot on the data compute node.
Callout number 3

GPU slot on the data compute node.
Callout number 4

I/O slot on the data compute node.
Callout number 5

I/O slot on the data compute node.

I/O slot configuration

The data compute node uses a specific slot numbering scheme that differs from standard server configurations. Understanding the slot layout is essential for proper cabling.

  • Slot 3: Reserved for GPU (not accessible for I/O cabling)

  • Slots 4 and 5: I/O slots used for network connections

    • Port a: Cluster network connections

    • Port b: Host network connections

  • Slots 1 and 2: Unpopulated and inaccessible for use

Network connections

Data compute nodes require two types of network connections to integrate with the AFX 1K storage system.

  • Host network connections

    Host network connections provide access to client data and enable the data compute nodes to process workloads. Each data compute node uses ports e4b and e5b for redundant connections to separate host network switches.

    Port assignments:

    • e4b: Connects to host network switch A

    • e5b: Connects to host network switch B

  • Cluster network connections

    Cluster network connections enable communication between data compute nodes and AFX 1K controller nodes within the storage cluster. Each data compute node uses ports e4a and e5a for redundant connections to separate cluster network switches.

    Port assignments:

    • e4a: Connects to cluster network switch A

    • e5a: Connects to cluster network switch B

Supported hardware components

The data compute nodes require specific cables and switches to ensure proper connectivity and performance with the AFX 1K storage system.

Data Compute Node Supported Switches Supported Cables

Data compute nodes (minimum of three required)

  • Cisco Nexus 9332D-GX2B (400GbE)

  • Cisco Nexus 9364D-GX2A (400GbE)

  • 400GbE QSFP-DD breakout to 4x100GbE QSFP56 cables for connections to data compute nodes:

    • 100GbE to data compute node cluster network ports (e4a, e5a)

    • 100GbE to data compute node host network ports (e4b, e5b)

  • RJ-45 cables for management connections
Note Breakout cables provide four 100GbE connections from each 400GbE switch port. Connect the 400GbE end to the switches and the 100GbE end to the data compute node I/O ports.

Cable orientation

When connecting cables to data compute nodes, proper cable orientation ensures reliable connections.

The cabling graphics in the installation procedures show arrow icons indicating the correct orientation (up or down) of the cable connector pull-tab when inserting a connector into a port. As you insert the connector, you should feel it click into place. If you do not feel it click, remove it, turn it over, and try again.

Cable pull tab direction

Caution Handle the delicate connector components carefully when clicking them into place.
What's next?

After reviewing the cabling configuration, cable the hardware for your data compute nodes.