Capacity Reservation Fleet concepts and planning - Amazon Elastic Compute Cloud
Plan a Capacity Reservation FleetTotal target capacityAllocation strategyInstance type weightInstance type priority

Capacity Reservation Fleet concepts and planning

The following information describes how to plan a Capacity Reservation Fleet and describes Capacity Reservation Fleet concepts including total target capacity, allocation strategy, instance type weight, and instance type priority.

Plan a Capacity Reservation Fleet

When planning your Capacity Reservation Fleet, we recommend that you do the following:

  1. Determine the amount of compute capacity that is needed by your workload.

  2. Decide on the instance types and Availability Zones that you want to use.

  3. Assign each instance type a priority based on your needs and preferences. For more information, see Instance type priority.

  4. Create a capacity weighting system that makes sense for your workload. Assign a weight to each instance type and determine your total target capacity. For more information, see Instance type weight and Total target capacity.

  5. Determine whether you need the Capacity Reservation indefinitely or only for a specific period of time.

Total target capacity

The total target capacity defines the total amount of compute capacity that the Capacity Reservation Fleet reserves. You specify the total target capacity when you create the Capacity Reservation Fleet. After the Fleet has been created, Amazon EC2 automatically creates Capacity Reservations to reserve capacity up to the total target capacity.

The number of instances for which the Capacity Reservation Fleet reserves capacity is determined by the total target capacity and the instance type weight that you specify for each instance type in the Capacity Reservation Fleet (total target capacity/instance type weight=number of instances).

You can assign a total target capacity based on units that are meaningful to your workload. For example, if your workload requires a certain number of vCPUs, you can assign the total target capacity based on the number of vCPUs required. If your workload requires 2048 vCPUs, specify a total target capacity of 2048 and then assign instance type weights based on the number of vCPUs provided by the instance types in the Fleet. For an example, see Instance type weight.

Allocation strategy

The allocation strategy for your Capacity Reservation Fleet determines how it fulfills your request for reserved capacity from the instance type specifications in the Capacity Reservation Fleet configuration.

Currently, only the prioritized allocation strategy is supported. With this strategy, the Capacity Reservation Fleet creates Capacity Reservations using the priorities that you have assigned to each of the instance type specifications in the Capacity Reservation Fleet configuration. Lower priority values indicate higher priority for use. For example, say you create a Capacity Reservation Fleet that uses the following instance types and priorities:

  • m4.16xlarge — priority = 1

  • m5.16xlarge — priority = 3

  • m5.24xlarge — priority = 2

The Fleet first attempts to create Capacity Reservations for m4.16xlarge. If Amazon EC2 has insufficient m4.16xlarge capacity, the Fleet attempts to create Capacity Reservations for m5.24xlarge. If Amazon EC2 has insufficient m5.24xlarge capacity, the Fleet creates Capacity Reservations for m5.16xlarge.

Instance type weight

The instance type weight is a weight that you assign to each instance type in the Capacity Reservation Fleet. The weight determines how many units of capacity each instance of that specific instance type counts toward the Fleet's total target capacity.

You can assign weights based on units that are meaningful to your workload. For example, if your workload requires a certain number of vCPUs, you can assign weights based on the number of vCPUs provided by each instance type in the Capacity Reservation Fleet. In this case, if you create a Capacity Reservation Fleet using m4.16xlarge and m5.24xlarge instances, you would assign weights that correspond to the number of vCPUs for each instance as follows:

  • m4.16xlarge — 64 vCPUs, weight = 64 units

  • m5.24xlarge — 96 vCPUs, weight = 96 units

The instance type weight determines the number of instances for which the Capacity Reservation Fleet reserves capacity. For example, if a Capacity Reservation Fleet with a total target capacity of 384 units uses the instance types and weights in the preceding example, the Fleet could reserve capacity for 6 m4.16xlarge instances (384 total target capacity/64 instance type weight=6 instances), or 4 m5.24xlarge instances (384 / 96 = 4).

If you do not assign instance type weights, or if you assign an instance type weight of 1, the total target capacity is based purely on instance count. For example, if a Capacity Reservation Fleet with a total target capacity of 384 units uses the instance types in the preceding example, but omits the weights or specifies a weight of 1 for both instance types, the Fleet could reserve capacity for either 384 m4.16xlarge instances or 384 m5.24xlarge instances.

Instance type priority

The instance type priority is a value that you assign to the instance types in the Fleet. The priorities are used to determine which of the instance types specified for the Fleet should be prioritized for use.

Lower priority values indicate a higher priority for use.