Lab: HCI + Compute Deployment

Objective

Deploy VergeOS in an HCI + Dedicated Compute configuration using the Terraform playground. You will provision a two-cluster topology — an HCI foundation cluster (controller + storage) and a dedicated compute-only cluster — then configure workload placement across clusters, validate independent compute scaling, and compare the operational model to pure HCI deployments.

Prerequisites

• Completed all prior modules (1–9)
• Completed the HCI Deployment Lab and UCI Deployment Lab
• Access to the vergeos-terraform-playground repository (cloned locally)
• Terraform CLI installed and configured
• A VergeOS environment or lab that supports nested deployments
• Familiarity with Terraform basics (init, plan, apply)

Difficulty

Intermediate — Requires understanding of VergeOS cluster architecture, multi-cluster networking, and basic Terraform usage

Estimated Time

1.5 hours

---

Background: HCI + Dedicated Compute Architecture

Before starting the lab, review the two-cluster model that defines HCI + Dedicated Compute:

Key principles:

• Cluster 1 (HCI) always includes Nodes 1 & 2 with controllers and Tier 0 storage. Optional Nodes 3–4 add storage and (optionally) compute capacity.
• Cluster 2 (Compute Only) contains nodes dedicated entirely to running workloads — no storage overhead, maximum resources for VMs.
• The Compute toggle on the HCI cluster controls whether HCI nodes can also run workloads alongside storage/control functions.
• All compute-node storage I/O traverses the core network to the HCI cluster, making inter-cluster bandwidth critical.

VMware Bridge

In VMware environments, a similar pattern involves running vSAN on a subset of hosts while adding compute-only hosts to the same vCenter cluster (or a separate cluster pointing to shared storage). VergeOS achieves this more cleanly with dedicated cluster roles — the HCI cluster owns all storage tiers, and compute-only clusters are explicitly defined with zero storage responsibility. There’s no need for separate vCenter constructs or shared datastores.

Nutanix Bridge

Nutanix supports compute-only nodes within an AHV cluster, but they still participate in the Nutanix CVM infrastructure. VergeOS takes a different approach: compute-only clusters are fully independent — no storage daemons, no CVM overhead. The compute nodes connect to the HCI cluster’s vSAN over the core fabric network, keeping the architecture cleaner and maximizing available resources for workloads.

---

Steps

Part 1: Review the HCI + Compute Topology

Understand the configuration before deploying.

1. In the terraform playground repository, navigate to the examples/ directory and identify the HCI + Compute .tfvars file (look for files referencing “hci-compute” or “hybrid” topologies)
2. Examine the variables and identify:
   • How many clusters are defined and their roles (HCI vs compute-only)
   • Node count and assignments per cluster
   • The Compute toggle setting on the HCI cluster — is it enabled or disabled?
   • Storage tier configuration (Tier 0 for metadata on controller nodes, Tier 1 for workload data)
   • Network configuration for inter-cluster communication
3. Compare this .tfvars file with the pure HCI configurations from the previous lab. Note the structural differences:
   • Additional cluster definition for compute-only nodes
   • Storage tier assignment — compute-only nodes have no storage tiers
   • Network bandwidth requirements between clusters
4. Review the deployment scenario documentation (docs/deployment-scenarios.md) for the HCI + Compute section

Part 2: Deploy the HCI + Compute Topology

Provision the two-cluster environment.

1. Run terraform init to initialize the provider (if not already done)
2. Run terraform plan -var-file=<hci-compute>.tfvars and carefully review the planned resources:
   • Verify two separate clusters will be created
   • Confirm node assignments match the expected topology
   • Check that storage tiers are only assigned to HCI cluster nodes
   • Validate network interfaces are configured for inter-cluster communication
3. Run terraform apply -var-file=<hci-compute>.tfvars to deploy
4. Log into the VergeOS UI and verify the deployment:
   • Clusters: Both clusters appear — one labeled HCI, one labeled Compute
   • Nodes: Each node is assigned to its correct cluster
   • Storage: vSAN storage pools exist only on the HCI cluster; compute-only nodes show no storage
   • Networking: Core fabric network connects both clusters; inter-cluster connectivity is established
   • Controllers: Controller VMs are running on Nodes 1 and 2 in the HCI cluster

Part 3: Configure Workload Placement

Practice placing workloads across the two-cluster topology.

1. Create a VM on the compute-only cluster:

   • In the VergeOS UI, create a new VM and select the compute-only cluster for placement
   • Assign CPU and memory resources
   • Attach a virtual disk — note that the storage is provisioned from the HCI cluster’s vSAN even though the VM runs on a compute-only node
   • Start the VM and verify it boots successfully

2. Create a VM on the HCI cluster (if Compute is enabled):

   • Create a second VM, this time placing it on the HCI cluster
   • Compare the resource availability between the two clusters
   • Note the difference: HCI nodes share resources between storage/control and compute, while compute-only nodes dedicate all resources to workloads

3. Test VM migration between clusters:

   • Attempt to migrate a VM from the compute-only cluster to the HCI cluster (and vice versa)
   • Observe how VergeOS handles cross-cluster migration
   • Document any constraints or considerations for cross-cluster workload movement

4. Monitor inter-cluster I/O:

   • Open the VergeOS dashboard and navigate to network monitoring
   • Observe the storage I/O traffic flowing from compute-only nodes to the HCI cluster
   • Note the bandwidth utilization on the core network — this is why inter-cluster bandwidth planning is critical

Part 4: Validate Independent Compute Scaling

Demonstrate the scaling advantage of the HCI + Compute model.

1. Review compute-only cluster capacity:

   • In the VergeOS UI, check the total CPU and memory available on the compute-only cluster
   • Compare this to the HCI cluster’s available compute resources (after storage/control overhead)
   • Document the effective compute capacity difference

2. Simulate a scale-out scenario:

   • Review the .tfvars file and identify how to add additional compute-only nodes
   • Modify the node count for the compute-only cluster (e.g., add 1–2 more nodes)
   • Run terraform plan to preview the change — note that only compute nodes are added; storage is unaffected
   • Apply the change and verify the new nodes join the compute-only cluster
   • Confirm that the HCI cluster is completely unchanged — no rebalancing, no storage disruption

3. Compare scaling models:

   Scaling Action	Pure HCI	HCI + Compute
   Add compute capacity	Must add full HCI node (with storage)	Add lightweight compute-only node
   Add storage capacity	Add HCI node or expand existing disks	Add node to HCI cluster only
   Scale independently	❌ Compute and storage coupled	✅ Compute scales independently
   Hardware flexibility	All nodes need storage-class hardware	Compute nodes optimized for workloads
   Operational complexity	Simple — single cluster	Moderate — two clusters, inter-cluster networking

Part 5: Explore the Compute Toggle

Understand the impact of the HCI cluster’s Compute setting.

1. Check the current Compute toggle state:

   • In the VergeOS UI, navigate to the HCI cluster settings
   • Identify whether the Compute toggle is currently enabled or disabled
   • If enabled, note which workloads (if any) are running on HCI nodes

2. Understand the two modes:

   Setting	Behavior	Best For
   Compute Enabled	HCI nodes run workloads alongside storage/control	Smaller deployments where maximizing utilization is preferred
   Compute Disabled	HCI cluster dedicated to storage and control only	Performance-sensitive environments where storage/compute isolation is preferred

3. Document your recommendation:

   • Based on the current deployment size, which Compute toggle setting would you recommend?
   • What factors would cause you to change the setting?
   • Note: Changing the Compute toggle may require a rolling restart of nodes in the HCI cluster — review the impact with VergeOS support before making this change in production

Part 6: Design Decision Exercise

Apply what you’ve learned to a real-world scenario.

1. Scenario: A customer currently runs a 4-node VergeOS HCI cluster. They need to add 50 new VMs for a development environment but don’t need additional storage. Their current storage utilization is only 40%, but CPU is at 75%.

2. Evaluate the options:

   • Option A: Add 2 more HCI nodes (6-node HCI cluster)
   • Option B: Add a 2-node compute-only cluster (4-node HCI + 2-node compute)
   • Option C: Migrate to full UCI architecture

3. For each option, document:

   • Hardware cost implications
   • Operational complexity change
   • Network requirements
   • Future scalability path
   • Your recommendation with justification

4. Bonus: Identify which terraform playground example most closely matches Option B, and list the .tfvars modifications needed to match the customer’s requirements

---

Cleanup

When finished with the lab:

1. Remove all test VMs created during the lab
2. Run terraform destroy to tear down the entire HCI + Compute deployment
3. Verify all resources have been cleaned up in the VergeOS UI

---

Verification

Your HCI + Compute deployment lab is complete when you can answer yes to all of the following:

• Successfully deployed a two-cluster HCI + Compute topology via Terraform
• Verified cluster roles (HCI vs compute-only), storage allocation, and networking in the VergeOS UI
• Created VMs on the compute-only cluster and confirmed storage was served from the HCI cluster
• Monitored inter-cluster storage I/O traffic on the core network
• Successfully scaled the compute-only cluster independently (added nodes without affecting storage)
• Documented the Compute toggle behavior and your recommendation for the deployment
• Completed the design decision exercise comparing HCI, HCI+Compute, and UCI options
• Cleaned up all lab resources with terraform destroy