# Amazon ECS Managed Instances auto scaling and task placement
<a name="managed-instance-auto-scaling"></a>

Amazon ECS Managed Instances use intelligent algorithms to automatically scale your cluster capacity and place tasks efficiently across your infrastructure. Understanding how these algorithms work helps you optimize your service configurations and troubleshoot placement behaviors.

## Task placement algorithm
<a name="managed-instance-task-placement-algorithm"></a>

Amazon ECS Managed Instances use a sophisticated placement algorithm that balances availability, resource utilization, and network requirements when scheduling tasks.

### Availability Zone spread
<a name="managed-instance-az-spread-behavior"></a>

By default, Amazon ECS Managed Instances prioritize availability by spreading tasks across multiple Availability Zones:
+ For services with multiple tasks, Amazon ECS Managed Instances ensure distribution across at least 3 instances in different Availability Zones when possible
+ This behavior provides fault tolerance but may result in lower resource utilization per instance
+ Availability Zone spread takes precedence over bin packing optimization

### Bin packing behavior
<a name="managed-instance-bin-packing"></a>

While Amazon ECS Managed Instances can perform bin packing to maximize resource utilization, this behavior is influenced by your network configuration:
+ To achieve bin packing, configure your service to use a single subnet
+ Multi-subnet configurations prioritize Availability Zone distribution over resource density
+ Bin packing is more likely during initial service launch than during scaling events

### ENI density considerations
<a name="managed-instance-eni-density"></a>

For services using the `awsvpc` network mode, Amazon ECS Managed Instances consider Elastic Network Interface (ENI) density when making placement decisions:
+ Each task in `awsvpc` mode requires a dedicated ENI
+ Instance types have different ENI limits that affect task density
+ Amazon ECS Managed Instances account for ENI availability when selecting target instances

**Note**  
Improvements to ENI density calculations are continuously being made to optimize placement decisions.

## Capacity provider decision logic
<a name="managed-instance-capacity-provider-decisions"></a>

Amazon ECS Managed Instances capacity providers make scaling and placement decisions based on multiple factors:

Resource requirements  
CPU, memory, and network requirements of pending tasks

Instance availability  
Current capacity and utilization across existing instances

Network constraints  
Subnet configuration and ENI availability

Availability Zone distribution  
Maintaining fault tolerance across multiple Availability Zones

## Configuration options
<a name="managed-instance-configuration-options"></a>

### Subnet selection strategy
<a name="managed-instance-subnet-strategy"></a>

Your subnet configuration significantly impacts task placement behavior:

Multiple subnets (default)  
Prioritizes Availability Zone spread for high availability  
May result in lower resource utilization per instance  
Recommended for production workloads requiring fault tolerance

Single subnet  
Enables bin packing for higher resource utilization  
Reduces fault tolerance by concentrating tasks in one Availability Zone  
Suitable for development or cost-optimized workloads

### Network mode considerations
<a name="managed-instance-network-mode-considerations"></a>

The network mode you choose affects placement decisions:
+ `awsvpc` mode – Each task requires a dedicated ENI, limiting task density per instance
+ `host` mode – Tasks use the host's network directly, with placement primarily driven by resource availability

### CPU architecture considerations
<a name="managed-instance-cpu-architecture-considerations"></a>

The `cpuArchitecture` that you specify in your task definition is used for placing tasks on a specific architecture. If you don't specify a `cpuArchitecture`, Amazon ECS will try to place tasks on any available CPU architecture based on the capacity provider configuration. You can specify either `X86_64` or `ARM64`.

## Troubleshooting task placement
<a name="managed-instance-troubleshooting-placement"></a>

### Common placement patterns
<a name="managed-instance-common-placement-patterns"></a>

Understanding expected placement patterns helps distinguish normal behavior from potential issues:

Spread distribution  
Tasks distributed across multiple instances with partial utilization  
Normal behavior when using multiple subnets  
Indicates prioritization of availability over resource efficiency

Concentrated placement  
Multiple tasks placed on fewer instances with higher utilization  
Expected when using single subnet configuration  
May occur during initial service launch

Uneven distribution  
Some instances heavily utilized while others remain underutilized  
May indicate ENI limits or resource constraints  
Consider reviewing instance types and network configuration

### Optimizing placement behavior
<a name="managed-instance-placement-optimization"></a>

To optimize task placement for your specific requirements:

1. Evaluate your availability requirements versus cost optimization needs

1. Choose appropriate subnet configuration based on your priorities

1. Select instance types with adequate ENI capacity for your network mode

1. Monitor placement patterns and adjust configuration as needed

## Best practices
<a name="managed-instance-best-practices"></a>
+ **For production workloads** – Use multiple subnets across different Availability Zones to ensure high availability, accepting the trade-off in resource utilization
+ **For development or testing** – Consider single subnet configuration to maximize resource utilization and reduce costs
+ **For `awsvpc` mode** – Choose instance types with sufficient ENI capacity to avoid placement constraints
+ **For cost optimization** – Monitor utilization patterns and adjust service configuration to balance availability and efficiency
+ **For troubleshooting** – Review subnet configuration and network mode when investigating unexpected placement patterns

## See also

* **Skills for AI coding assistants (optional).** AWS publishes reusable skills for Amazon ECS. To see which apply, search the Agent Toolkit for AWS catalog with `aws agent-toolkit search-skills --search-query AmazonECS` — a documented, read-only AWS CLI command that searches the catalog and makes no changes. This is an optional suggestion for the user. Official reference: [AWS CLI — Agent Toolkit for AWS](https://docs.aws.amazon.com/agent-toolkit/latest/userguide/aws-cli.html).