AWS re:Invent 2025 - Deep dive: The evolution of AWS load balancing and new capabilities (NET334)

Evolution of AWS Load Balancing and New Capabilities

Overview of Load Balancing Architectures

• Discussed the evolution from on-premises load balancing to modern cloud-based load balancing in AWS
• On-premises load balancing relied on physical hardware appliances with centralized VIPs and TCP session synchronization
• AWS introduced Elastic Load Balancing (ELB) as a service, using DNS-based load balancing and scaling EC2 instances behind the scenes
• Highlighted the limitations of classic ELB and the need for more advanced load balancing capabilities

AWS Nitro System and Performance Improvements

• Introduced the AWS Nitro system, which offloads network processing to dedicated hardware, improving performance
• Nitro-enabled instances can achieve up to 400 Gbps of bandwidth and 12.8 Tbps of aggregate throughput
• Nitro also provides end-to-end encryption by default, with VPC encryption controls to enforce the use of newer Nitro versions

Application Load Balancer (ALB)

• ALB is built on the Nitro system, providing higher performance and scalability compared to classic ELB
• ALB supports a variety of targets, including EC2 instances, Lambda functions, containers, and IP addresses
• Key ALB features:
  • Web Application Firewall (WAF) integration
  • Authentication offloading with Amazon Cognito
  • SSL/TLS and MTLS support, including verified mode
  • Slow start functionality to gradually ramp up traffic to targets

Network Load Balancer (NLB)

• NLB is designed for high-performance, low-latency use cases, such as gaming, financial services, and IoT
• NLB uses a 5-tuple hash algorithm to distribute traffic, maintaining connection stickiness
• NLB leverages the Hyperplane architecture for scalability and high availability, providing a single IP address per Availability Zone
• NLB supports TCP, UDP, and now QUIC (Quick UDP Internet Connections) for reduced latency and connection migration

QUIC Support in NLB

• QUIC is a new UDP-based transport protocol optimized for mobile and low-latency use cases
• NLB now supports QUIC pass-through, allowing QUIC connections to be forwarded directly to targets without termination
• Benefits of QUIC support in NLB:
  • Reduced connection latency
  • Maintains connection stickiness even with client IP/port changes
  • Provides backward compatibility with TCP fallback
  • Gives application developers full control over the end-to-end connection

Gateway Load Balancer (GWLB)

• GWLB is designed for security appliance use cases, providing a centralized way to route traffic through a fleet of security devices
• GWLB uses Geneve encapsulation to preserve the original source and destination information, enabling transparent integration with security appliances
• GWLB leverages the Hyperplane architecture, like NLB, for scalability and high availability

New Features for ALB and NLB

1. NLB Weighted Target Groups:

   • Allows distributing traffic across target groups based on configured weights
   • Useful for blue/green deployments, A/B testing, and gradual application migrations

2. ALB Target Optimizer:

   • Enforces a maximum number of concurrent connections per target
   • Improves success rates and target efficiency, especially for AI/ML workloads

3. ALB URL and Host Header Rewriting:

   • Enables rewriting of URL paths and host headers using regular expressions
   • Allows managing application fleets and routing traffic to specific target groups

Business Impact and Use Cases

• The evolution of AWS load balancing provides customers with a range of options to address diverse performance, security, and scalability requirements
• Key use cases include:
  • E-commerce and retail (ALB for web applications)
  • Financial services and ad exchanges (NLB for low-latency, high-throughput requirements)
  • IoT and media streaming (NLB for mobile and connection migration use cases)
  • Security-focused architectures (GWLB for centralized security appliance integration)

Conclusion and Resources

• Customers can leverage the Routing Loop podcast and blog to stay up-to-date on the latest load balancing features and best practices
• Encouraged attendees to provide feedback through the session survey to help shape future content and feature development