Predictika · NetApp AFX cluster design

An AI agent's AFX cluster design, checked before it is quoted

An SE, a partner or an AI agent proposes a NetApp AFX cluster for an AI data platform: which AFX node model, how many nodes and shelves, which Cisco Nexus cluster switch pair, and whether the AI Data Engine and its data compute nodes are deployed. Before the design goes to a quote or an order, the proposal goes to the Predictika validator. It computes the 400GbE ports each cluster switch needs from the nodes and shelves, compares them with the ports the chosen switch has, checks the node, shelf and data compute node limits NetApp documents, compares the port count the proposal states with the computed one, and returns every reason in one answer. Hard rules block; configurations NetApp documents differently, or sizes in its Hardware Universe, route to a person. This checks that a design is legal, not how it performs, and nothing is priced. This is an illustrative model built from NetApp's public documentation. NetApp is not involved with it and has not endorsed it.

Scope: NetApp AFX 1K and AFX 2K clusters on Cisco Nexus 9332D-GX2B, 9364D-GX2A or 9808 cluster switches, with or without the AI Data Engine. See everything this validator covers

This validator checks one kind of proposal: the node, shelf and cluster switch layout of a NetApp AFX cluster, as described in NetApp's AFX hardware and AI Data Engine hardware documentation. It handles any proposal that picks from these models and switches, not only the scenarios on this page. It does not size capacity or performance, and it does not cover drives, the data-facing network, configuration or pricing.

  • Covered: cluster limits — an even node count (HA pairs), a maximum of 32 nodes and at least 2 shelves; a node-to-shelf ratio outside the documented 4:1 to 8:3 goes to approval.
  • Covered: switch ports — the 400GbE ports each cluster switch needs: 1 per AFX 1K node and 3 per AFX 2K node, plus 2 per shelf, against the 30 usable ports of the Nexus 9332D-GX2B and the 62 of the 9364D-GX2A. The Nexus 9808 is sized in NetApp's Hardware Universe and always goes to approval.
  • Covered: the AI Data Engine — a minimum of 3 data compute nodes when it is deployed, and the client and cluster switches the documentation lists.
  • Covered: the proposal itself — the switch port count the proposal states is compared with the computed one; a mismatch blocks.
  • Not covered — capacity and drive choices, performance, the data-facing network, cabling and breakout details, software versions, configuration and price. Other areas can be added.
  • Only stated rules are modeled — the documents do not state how many drives a shelf holds or a minimum node count, so neither is checked.
The AI design agentProposes the AFX cluster from the customer's requirements and says it is ready to quote.
Predictika validatorChecks every rule at once and returns the verdict, each finding, and the computed figures.

8 scenarios in 4 groups · each group runs from the agent’s first attempt to the outcome · click one to switch

What the agent proposes

The AI design agent says

Illustrative agent statement. The validator receives the structured request below.

Proposing agent: AI design agent (scripted)

Before sending, the agent must call Recorded run

Predictika validator


      

        
        
        
      

      
Waiting for the call.

      
    

  

  
  

    
Where rules-based CPQ gets it wrong
Click to expand
    
Our assessment of what a typical rules-based CPQ (product rules, price rules and validation rules, as in Salesforce CPQ) does with the same cases. It is a judgment about effort and reliability, not a claim that nothing can be built.

    
What the validator caughtTypical rules-based CPQPredictika
Switch ports that depend on the node model, the shelves and the switchOnly with workarounds
Each AFX 1K node takes one 400GbE switch port and each AFX 2K node three, each shelf two, and the switch has 30 or 62 usable ports. That is arithmetic across three catalogs, and a configurator passes a cluster that fits the nodes but not the switch.		The ports are computed from the nodes, the shelves and the model on every call and compared with the usable ports of the chosen switch; the shortfall comes back with both numbers.
Cluster limits that interact: HA pairs, the 32-node maximum, shelf count and ratioOnly with care
Even node counts, a maximum and a shelf minimum are simple alone, but the node-to-shelf ratio depends on both counts and has a boundary at each end.		All four are checked together; a ratio outside the documented range goes to a person rather than being blocked or ignored.
Rules for an attached product, the AI Data EngineOnly with care
The AI Data Engine adds a minimum number of data compute nodes and its own list of switches. A cluster rule set rarely knows about the attached product.		The data compute node minimum and the listed switches are checked when the AI Data Engine is on, and not when it is off.
Hard rules versus a switch NetApp sizes elsewhere, and figures the proposal itself statesOnly with workarounds
Approval chains handle one threshold well, but here some findings block, the Nexus 9808 is sized in another tool, and the proposal's own stated port count must be checked against the computed one.		Each finding is labeled block or approval, and the stated port count travels with the proposal and is compared on every call.

  

  

    
What this validator catches: findings from its test cases
Click to expand
    
Every rule is checked on every call, in any order the choices arrive; one problem never hides another. Blocking findings stop the design; approval findings need a person to sign off; everything else is computed. The wording shown is an example from the test runs; the numbers change with the proposal.

    
1Fewer than 2 shelves (blocks): An AFX cluster needs at least 2 shelves; the proposal has 1.
2Fewer than 3 data compute nodes (blocks): The AI Data Engine needs a minimum of 3 data compute nodes; the proposal has 2.
3More than 32 nodes (blocks): An AFX cluster has a maximum of 32 nodes; the proposal has 34.
4Node count is not even (blocks): An AFX cluster is built from HA pairs, so the node count must be even; the proposal has 23 nodes.
5Not enough switch ports (blocks): Each cluster switch needs 35 x 400GbE ports (23 nodes x 1 + 6 shelves x 2) but the Nexus 9332D-GX2B has 30 usable ports.
6Stated switch ports do not match (blocks): Stated switch ports do not match: the agent stated 28; the validator computes 35 x 400GbE ports per switch.
7Nexus 9808 is sized in the Hardware Universe (needs approval): The Nexus 9808 is for configurations above 64 ports and is sized in the NetApp Hardware Universe; this validator does not size it, so confirm the switch ports there.
8Outside the node-to-shelf ratio (needs approval): 16 nodes with 7 shelves is outside the 4:1 to 8:3 node-to-shelf ratio NetApp documents; confirm the shelf count with NetApp.
9Switch not listed for the AI Data Engine (needs approval): The AI Data Engine documentation lists the Nexus 9332D-GX2B or 9364D-GX2A for client and cluster switches; Nexus 9808 is not listed there.

  

  

    
Scenarios and results
Click to expand
    
The recorded result of the validator for each scenario on this page: the request the AI agent sent, the verdict, and every finding, in the order they appear above.

    
1. Nodes, ports and switches: 23 nodes, 6 shelves on a 9332D pair, AI Data Engine with 2 compute nodes

    
The agent's request. Please check the AFX cluster for Northwind AI Labs: 23 AFX 1K nodes and 6 shelves on a Nexus 9332D-GX2B cluster switch pair, with the AI Data Engine and 2 data compute nodes, stated as 28 400GbE ports needed on each switch. It is ready to quote.

    
Verdict: Blocked.

    
    
      
  • Node count is not even. An AFX cluster is built from HA pairs, so the node count must be even; the proposal has 23 nodes.
    • 
      
  • Not enough switch ports. Each cluster switch needs 35 x 400GbE ports (23 nodes x 1 + 6 shelves x 2) but the Nexus 9332D-GX2B has 30 usable ports.
    • 
      
  • Fewer than 3 data compute nodes. The AI Data Engine needs a minimum of 3 data compute nodes; the proposal has 2.
    • 
      
  • Stated switch ports do not match. Stated switch ports do not match: the agent stated 28; the validator computes 35 x 400GbE ports per switch.
    • 
    

    
2. Nodes, ports and switches: Even node count and 3 compute nodes, still the 9332D pair

    
The agent's request. Please check the AFX cluster for Northwind AI Labs: 22 AFX 1K nodes and 6 shelves on a Nexus 9332D-GX2B cluster switch pair, with the AI Data Engine and 3 data compute nodes, stated as 34 400GbE ports needed on each switch. It is ready to quote.

    
Verdict: Blocked.

    
    
      
  • Not enough switch ports. Each cluster switch needs 34 x 400GbE ports (22 nodes x 1 + 6 shelves x 2) but the Nexus 9332D-GX2B has 30 usable ports.
    • 
    

    
3. Nodes, ports and switches: The same cluster on a 9364D pair

    
The agent's request. Please check the AFX cluster for Northwind AI Labs: 22 AFX 1K nodes and 6 shelves on a Nexus 9364D-GX2A cluster switch pair, with the AI Data Engine and 3 data compute nodes, stated as 34 400GbE ports needed on each switch. It is ready to quote.

    
Verdict: Allowed.

    
4. AFX 2K backend ports: 12 AFX 2K nodes and 3 shelves on a 9332D pair

    
The agent's request. Please check the AFX cluster for Fjord Analytics: 12 AFX 2K nodes and 3 shelves on a Nexus 9332D-GX2B cluster switch pair, with the AI Data Engine and 3 data compute nodes, stated as 42 400GbE ports needed on each switch. It is ready to quote.

    
Verdict: Blocked.

    
    
      
  • Not enough switch ports. Each cluster switch needs 42 x 400GbE ports (12 nodes x 3 + 3 shelves x 2) but the Nexus 9332D-GX2B has 30 usable ports.
    • 
    

    
5. AFX 2K backend ports: The same cluster on a 9364D pair

    
The agent's request. Please check the AFX cluster for Fjord Analytics: 12 AFX 2K nodes and 3 shelves on a Nexus 9364D-GX2A cluster switch pair, with the AI Data Engine and 3 data compute nodes, stated as 42 400GbE ports needed on each switch. It is ready to quote.

    
Verdict: Allowed.

    
6. Beyond the documented configurations: 16 nodes, 7 shelves, Nexus 9808 pair, AI Data Engine on

    
The agent's request. Please check the AFX cluster for Skerry Media: 16 AFX 1K nodes and 7 shelves on a Nexus 9808 cluster switch pair, with the AI Data Engine and 3 data compute nodes, stated as 30 400GbE ports needed on each switch. It is ready to quote.

    
Verdict: Needs approval.

    
    
      
  • Outside the node-to-shelf ratio. 16 nodes with 7 shelves is outside the 4:1 to 8:3 node-to-shelf ratio NetApp documents; confirm the shelf count with NetApp.
    • 
      
  • Nexus 9808 is sized in the Hardware Universe. The Nexus 9808 is for configurations above 64 ports and is sized in the NetApp Hardware Universe; this validator does not size it, so confirm the switch ports there.
    • 
      
  • Switch not listed for the AI Data Engine. The AI Data Engine documentation lists the Nexus 9332D-GX2B or 9364D-GX2A for client and cluster switches; Nexus 9808 is not listed there.
    • 
    

    
7. Cluster limits: 34 nodes against the 32-node maximum

    
The agent's request. Please check the AFX cluster for Tundra Research: 34 AFX 1K nodes and 10 shelves on a Nexus 9364D-GX2A cluster switch pair, with the AI Data Engine and 3 data compute nodes, stated as 54 400GbE ports needed on each switch. It is ready to quote.

    
Verdict: Blocked.

    
    
      
  • More than 32 nodes. An AFX cluster has a maximum of 32 nodes; the proposal has 34.
    • 
    

    
8. Cluster limits: 32 nodes, the maximum

    
The agent's request. Please check the AFX cluster for Tundra Research: 32 AFX 1K nodes and 10 shelves on a Nexus 9364D-GX2A cluster switch pair, with the AI Data Engine and 3 data compute nodes, stated as 52 400GbE ports needed on each switch. It is ready to quote.

    
Verdict: Allowed.

  

  

    
Narration transcript
Click to expand
    
The text of the narrated walkthrough, step by step.

    
    
      
  1. Welcome to a demo of Predictika's Validator. An SE, a partner or an AI agent proposes a NetApp AFX cluster for an AI data platform: which AFX node model, how many nodes and shelves, which Cisco Nexus cluster switch pair, and more. Before the design goes to a quote or an order, the proposal goes to the Predictika validator. It computes the 400GbE ports each cluster switch needs from the nodes and shelves, compares them with the ports the chosen switch has, checks the node, shelf and data compute node limits NetApp documents, compares the port count the proposal states with the computed one, and returns every reason in one answer. Hard rules block; configurations NetApp documents differently, or sizes in its Hardware Universe, route to a person. This checks that a design is legal, not how it performs, and nothing is priced. This is an illustrative model built from NetApp's public documentation. NetApp is not involved with it and has not endorsed it.
    2. 
      
  2. I will start with the first request by the agent.
    3. 
      
  3. The agent says. Please check the AFX cluster for Northwind AI Labs: 23 AFX 1K nodes and 6 shelves on a Nexus 9332D-GX2B cluster switch pair. It is ready to quote. The rest of the request is on screen.
    4. 
      
  4. Before it goes out, the agent calls the validator.
    5. 
      
  5. Blocked, with four findings, all in one answer. An AFX cluster is built from HA pairs, so the node count must be even; the proposal has 23 nodes. Each cluster switch needs 35 x 400GbE ports (23 nodes x 1 + 6 shelves x 2) but the Nexus 9332D-GX2B has 30 usable ports. The AI Data Engine needs a minimum of 3 data compute nodes; the proposal has 2. The other one is on screen.
    6. 
      
  6. The validator returned four blocking findings in one answer. A check that reads one field at a time would pass most of this; it shows only when the choices in the design are checked against each other.
    7. 
      
  7. What this shows: for switch ports that depend on the node model, the shelves and the switch, a rules-based application can do this only with workarounds.
    8. 
      
  8. Now the agent resubmits.
    9. 
      
  9. The corrected request. It changes the AFX nodes to 22, the data compute nodes to 3 and the stated 400GbE ports per switch to 34.
    10. 
      
  10. It calls the validator again.
    11. 
      
  11. Still blocked, with one finding, all in one answer. Each cluster switch needs 34 x 400GbE ports (22 nodes x 1 + 6 shelves x 2) but the Nexus 9332D-GX2B has 30 usable ports.
    12. 
      
  12. Three findings cleared. Still blocking: not enough switch ports.
    13. 
      
  13. The agent resubmits once more.
    14. 
      
  14. The revised request. It changes the cluster switch pair to Nexus 9364D-GX2A.
    15. 
      
  15. It calls the validator once more.
    16. 
      
  16. This time the validator clears it.
    17. 
      
  17. The validator also returns the line-item quote. It starts with AFX 1K nodes and NVMe SSD shelves; the other line is on screen.
    18. 
      
  18. The earlier finding is cleared, and the validator's figure matches the design.
    19. 
      
  19. Four problems caught in one call, fixed in two passes, and a clean design. That is the value of Predictika's validator: catching mistakes before they reach the build. Now you can try the other scenarios yourself.
    20.