Ollama on Aruba Cloud
Deploy Ollama — a local LLM inference platform — on Aruba Cloud using Terraform and cloud-init. Runs large language models (LLama, Mistral, Gemma, etc.) directly on CPU without requiring a GPU.
Provider version: arubacloud/arubacloud
~> 1.0| Terraform: ≥ 1.9
Introduction
Ollama makes running LLMs locally as simple as ollama run llama3.2. It provides an OpenAI-compatible REST API, enabling drop-in replacement for OpenAI in existing applications. This example provisions:
- Ollama installed via the official install script as a systemd service
- REST API bound to all interfaces (port 11434), restricted to
api_cidr - Optional model pre-pulling at bootstrap time
Resource warning: LLMs are memory-intensive. A 7B model requires ~4–6 GB free RAM, a 13B model ~8–10 GB, and a 70B model ~40+ GB. Plan your
vm_flavoraccordingly.
Architecture Overview
Infrastructure Created
| Resource | Name pattern | Description |
|---|---|---|
arubacloud_project | ollama-prod | Project container |
arubacloud_vpc | ollama-prod-vpc | Virtual Private Cloud |
arubacloud_subnet | ollama-prod-subnet | Basic subnet |
arubacloud_securitygroup | ollama-prod-vm-sg | Security group |
arubacloud_securityrule | ollama-prod-vm-ssh | SSH ingress |
arubacloud_securityrule | ollama-prod-vm-api | Ollama API TCP 11434 |
arubacloud_elasticip | ollama-prod-vm-eip | VM public IP |
arubacloud_blockstorage | ollama-prod-boot | 100 GB boot disk (Performance) |
arubacloud_keypair | ollama-prod-keypair | SSH public key |
arubacloud_cloudserver | ollama-prod-vm | CloudServer VM |
Estimated Monthly Cost
| Resource | Spec | Est. cost/mo |
|---|---|---|
| CloudServer VM | CSO8A16 — 8 vCPU / 16 GB | ~€55 |
| Boot disk | 100 GB Performance | ~€15 |
| Elastic IP | — | ~€3 |
| Total | ~€73/mo |
Larger models require upgrading to a VM with more RAM (and correspondingly higher cost).
Requirements
- Terraform ≥ 1.9
- ArubaCloud Terraform Provider
~> 1.0 - An ArubaCloud account with OAuth2 API credentials
- An SSH key pair
Variables
Required
| Variable | Description |
|---|---|
arubacloud_client_id | ArubaCloud OAuth2 client ID |
arubacloud_client_secret | ArubaCloud OAuth2 client secret |
ssh_public_key | SSH public key content |
Optional
| Variable | Default | Description |
|---|---|---|
app_name | "ollama" | Short name used in all resource names |
environment | "prod" | Environment label |
location | "ITBG-Bergamo" | ArubaCloud region |
zone | "ITBG-1" | Availability zone |
billing_period | "Hour" | "Hour" or "Month" |
vm_flavor | "CSO8A16" | CloudServer flavor |
vm_image | "LU22-001" | Boot disk image (Ubuntu 22.04 LTS) |
vm_disk_size_gb | 100 | Boot disk size in GB (min 50 GB) |
ssh_cidr | "0.0.0.0/0" | CIDR for SSH |
api_cidr | "0.0.0.0/0" | CIDR for REST API port 11434 — restrict to your app servers |
preload_models | [] | Models to pull at bootstrap (e.g. ["llama3.2"]) |
Outputs
| Output | Description |
|---|---|
ollama_url | Ollama REST API URL |
vm_public_ip | Public IP address of the VM |
ssh_command | SSH command to connect to the VM |
health_check | curl command to list available models |
Deployment Instructions
1. Clone and navigate
git clone https://github.com/arubacloud/terraform-arubacloud-examples.git
cd terraform-arubacloud-examples/ollama
2. Configure variables
cp terraform.tfvars.example terraform.tfvars
Restrict API access and optionally pre-pull models:
api_cidr = "10.0.0.0/8" # your application server CIDR
preload_models = ["llama3.2", "nomic-embed-text"]
3. Deploy
terraform init
terraform plan
terraform apply
Bootstrap takes 2–5 minutes (plus model pull time — a 7B model is ~4 GB).
4. Verify
curl http://$(terraform output -raw vm_public_ip):11434/api/tags
5. Run a model
ssh ubuntu@$(terraform output -raw vm_public_ip)
ollama run llama3.2
Available Models
# Browse all available models
ollama list
# Popular models
ollama pull llama3.2 # Meta Llama 3.2 3B — fast, low RAM
ollama pull llama3.1 # Meta Llama 3.1 8B — balanced
ollama pull mistral # Mistral 7B
ollama pull codellama # Code-optimised LLaMA
ollama pull nomic-embed-text # Text embedding model (for RAG)
ollama pull gemma2 # Google Gemma 2 9B
Security Recommendations
- Always restrict
api_cidr. Ollama has no built-in authentication. An open port 11434 allows anyone to run models and exhaust your server resources.