Skip to content

cost-estimation.md

Latest commit

 

History

History
95 lines (75 loc) · 4.22 KB

cost-estimation.md

File metadata and controls

95 lines (75 loc) · 4.22 KB
Raw

Cost Estimation

LabLink's AWS costs depend on your region, instance types, how long the allocator and client VMs run, attached storage, public IPv4 addresses, data transfer, and optional DNS or load balancing. Check the AWS Pricing Calculator for a current estimate before deploying.

Get a deployment estimate

For a CLI deployment, run lablink status. Its Cost Estimate (daily) section uses the AWS Pricing API when available and a built-in fallback table otherwise. It includes the allocator, EBS, optional ALB and Route 53 costs, and client VMs.

The template repository also has a script. Run it from the root of your template checkout; it reads lablink-infrastructure/config/config.yaml:

./scripts/estimate-costs.sh

The script requires jq; the AWS CLI enables live price lookups. Its current allocator EBS estimate assumes a 20 GiB root volume, while the template does not set root_block_device and uses the selected AMI's default. Check the deployed volume size before relying on that line of the estimate.

Compute

The template runs one t3.large allocator. Its instance type is fixed in lablink-infrastructure/main.tf; machine.machine_type selects the client VM type. The allocator accrues EC2 charges while running. Client EC2 usage scales with the number of machines and their running hours:

client compute cost = client count × hours running × hourly instance price

Destroy client VMs when a workshop ends. Stopping an EC2 instance ends its compute usage charge, but its EBS volumes still accrue storage charges; see AWS's instance lifecycle guide.

GPU Instance Types

The CLI wizard offers these GPU client types; config.yaml can specify any EC2 instance type supported in your region. Use the AWS Pricing Calculator for current regional prices.

Type GPU vCPUs RAM GPU memory
g4dn.xlarge T4 4 16 GiB 16 GiB
g4dn.2xlarge T4 8 32 GiB 16 GiB
g5.xlarge A10G 4 16 GiB 24 GiB
g5.2xlarge A10G 8 32 GiB 24 GiB
p3.2xlarge V100 8 61 GiB 16 GiB

The CPU-only wizard choices are t3.large, t3.xlarge, t3.2xlarge, m5.xlarge, and m5.2xlarge. LabLink does not currently provision Spot client instances.

Storage and networking

  • EBS: The template sets client root volumes to 80 GiB. It leaves the allocator root volume at the AMI default. EBS storage can keep accruing charges while an instance is stopped.
  • Public IPv4: AWS currently charges $0.005 per address-hour for both associated and idle public IPv4 addresses, including Elastic IPs. One continuously allocated address is about $3.65 for a 730-hour month. Check Amazon VPC pricing for the current rate and any applicable credits.
  • State storage: The S3 state bucket and DynamoDB lock table remain after a deployment is destroyed. S3 versioning retains old state versions until you manage them.
  • Optional services: Route 53 hosted zones and DNS queries, an ALB for ssl.provider: acm, and internet data transfer can add charges. The template does not create S3 database backups or CloudWatch log storage as part of a standard deployment.

Track and limit spending

Use AWS Cost Explorer or a budget in the Billing console to track actual charges. Template resources are tagged with Project=<deployment_name>, Environment=<environment>, and ManagedBy=terraform; filter by those values to inspect a deployment. The template's tag names are fixed in its OpenTofu files.

AWS Free Tier terms depend on when the account was created and its plan. Accounts created on or after July 15, 2025 use the newer credit-based program; see the AWS Free Tier guide and your account's Billing console. Do not assume a t3.large allocator or GPU client is covered by an older micro-instance allowance.

For deployment teardown, see Managing Deployments or Template Repository Deployment.