AWS re:Invent 2025 - Navigate cloud compute with Fargate and ECS Managed Instances (CNS342)

Navigating Cloud Compute with AWS Fargate and ECS Managed Instances

Overview of Amazon ECS

• ECS is a container orchestration service built for use on AWS
• Key principles:
  • Handles operational tasks so customers can focus on applications
  • Scales to support workloads of any size
  • Only charges for the resources used, not the orchestration service itself
  • Embeds AWS security and operational best practices
• ECS is widely adopted, with customers launching over 3 billion tasks per week
• Many AWS services and Amazon.com run on ECS, including 18 million Fargate tasks per day during Prime Day

Capacity Provisioning Options

• AWS provides a range of compute options for ECS:
  • Fargate: Fully managed compute, customers only specify CPU/memory requirements
  • EC2: Customers have full control over instance types and configuration
  • ECS Managed Instances: Balanced approach with managed compute but more control
• Fargate benefits:
  • Simplified operations, no compute management
  • Task isolation with single-use EC2 instances
  • Only pay for runtime, no charges when tasks are stopped
• EC2 benefits:
  • Full control over instance types and configurations
  • Ability to leverage EC2 purchasing models like reserved instances
• ECS Managed Instances:
  • EC2 instances provisioned and managed by AWS in customer accounts
  • Customers can specify instance type attributes but not control instances directly
  • Faster scaling, continuous cost optimization, and support for privileged containers

Security and Compliance

• Shared responsibility model:
  • AWS manages the ECS control plane, foundational services, and global infrastructure
  • Customers manage application-level security, load balancing, security groups, etc.
• Fargate provides strong isolation with single-use EC2 instances
• ECS Managed Instances leverage Bottlerocket OS and other AWS security best practices
• Compliance is simplified, as AWS provides evidence for many compliance requirements

Cost Optimization Strategies

• Hardware selection:
  • Fargate allows choosing CPU architecture (x86 or Graviton)
  • Managed Instances enable selecting optimal instance types for workloads
• Purchasing options:
  • On-Demand, Savings Plans, and Spot Instances
  • Fargate provides 100% utilization, no need to manage unused capacity
  • Managed Instances can consolidate tasks on underutilized instances
• Other optimizations:
  • Fargate's container image caching and lazy loading
  • Managed Instances' container image caching and active capacity management

Customer Journey: Cubert Research and Technologies (QRT)

• QRT is a global systematic asset manager that adopted ECS and Fargate
• Initial architecture used Python-based "Quant Computational Units" (QCUs) and "Head Nodes" for aggregation and caching
• Scaled to tens of thousands of services across multiple accounts and VPCs
• Evolved architecture to leverage multi-account, multi-region deployment for resilience and flexibility
• Exploring use of ECS Managed Instances to support GPU workloads and memory-intensive applications

Key Takeaways

• ECS provides a range of compute options to match different customer needs and workload requirements
• Fargate offers a fully managed experience, while Managed Instances balance control and simplicity
• Security, compliance, and cost optimization are core strengths of the ECS platform
• Customers like QRT have successfully scaled mission-critical workloads on ECS and Fargate