Porter supports intelligent cost optimization for node groups, allowing you to automatically optimize resource allocation and reduce costs while maintaining performance.

​
How Node Groups Work

By default, node groups in Porter use a fixed instance type approach. When you create a node group, you select a specific machine type (e.g., t3.xlarge with 4 CPU, 16GB RAM) and Porter scales by adding or removing instances of that exact type. While this works well for predictable workloads, it often leads to resource fragmentation.

Let’s look at an example. Say you’re running a web application with 8 replicas, each needing:

  • 2.5 CPU cores
  • 4GB RAM
  • Total requirements: 20 CPU cores and 32GB RAM

With fixed instance types, you might choose t3.xlarge instances (4 CPU, 16GB RAM each). You would need 5 instances to get enough CPU (20 cores needed), which provides 80GB of RAM - far more than the 32GB you need. At $0.1664 per hour for each instance ($0.832 total per hour), you’re paying for 48GB of unused RAM - a 60% waste in memory.

With cost optimization enabled, Porter can intelligently bin pack your workloads using a combination of instance types. Using 3 t3.large instances (2 CPU, 8GB RAM each) at $0.0832/hour ($0.2496 total) and 4 t3.medium instances (2 CPU, 4GB RAM each) at $0.0416/hour ($0.1664 total), Porter achieves the exact capacity needed (20 CPU cores and 32GB RAM) for $0.416/hour - a 50% reduction in cost with the same performance. The machines are reshuffled on a cadence to ensure the best fit as the set of workloads changes.

​
Enabling and Configuring Cost Optimization

To enable cost optimization for a node group:

  1. Navigate to your cluster’s infrastructure settings
  2. Under node groups, find the node group you want to optimize
  3. Click “Enable Cost Optimization” in the top right corner
  4. Set your maximum CPU cores limit to prevent unexpected scaling. This helps prevent unexpected cost increases by setting a cap.

Cost optimization configuration for a node group

​
Best Practices

Health Checks Required: For production applications, ensure proper health checks are configured before scheduling them on cost-optimized node groups. This ensures your applications can be safely rescheduled on new nodes without causing any disruption as nodes are reshuffled.

​
Limitations

The following node group configurations should continue using fixed instance types until we support cost optimization for them:

  • GPU instances (e.g., instances with NVIDIA GPUs)
  • Spot instances
  • Instances in public subnets
  • Instances with specialized hardware requirements