Great deep dive! I recently migrated a legacy app from a CLB to an ALB specifically for the path-based routing, and the difference in managing our microservices was night and day. Your explanation of the NLB's static IPs also solved a puzzle we had with a third-party firewall whitelist.

Great deep dive. I recently migrated a legacy app from a CLB to an NLB for its TCP-preserving features, and your explanation of the NLB's flow stickiness versus the ALB's request-based routing perfectly captures the key architectural decision we had to make.

Great overview of the core types. For the NLB's handling of long-lived TCP connections (like WebSockets), in your experience, does its static IP and flow stickiness ever complicate blue/green deployments compared to the ALB's more dynamic target group routing?

Great deep dive. I recently migrated a legacy app from a CLB to an NLB to handle a spike in TCP traffic, and the performance difference in latency was immediately noticeable, just as you outlined. Your explanation of the NLB's flow hash routing versus the ALB's content-based routing perfectly captures that key architectural decision point.

Great breakdown of the core types. For the NLB's handling of long-lived TCP connections (like WebSockets), do you have any best practices for managing client affinity or ensuring even distribution when connection counts per client vary widely?

Great breakdown of the three types. For a modern microservices architecture using ECS Fargate, would you still consider the CLB's simplicity for internal services, or is ALB's path-based routing always the preferred starting point?

Great deep dive. I recently migrated a legacy app from a CLB to an NLB for its TCP-preserving features, and the performance gain for our WebSocket connections was exactly as you described—night and day. Your breakdown of the "amateur hour" single-server trap is spot on.

Great breakdown. I made the classic mistake of using a CLB for a modern microservices app a few years ago; switching to ALB for its path-based routing was a game-changer for our team's deployment strategy.

While it's true that the Classic Load Balancer (CLB) has significant limitations compared to newer options like the Application Load Balancer (ALB), dismissing it entirely overlooks scenarios where it might still be useful. For legacy systems that require minimal configuration and basic health checks without the complexities of modern architectures, the CLB can still function adequately, particularly if the application demand is manageable and straightforward.

It's worth noting that the Application Load Balancer (ALB) supports features like WebSocket and HTTP/2, making it a better fit for modern applications requiring real-time communication. Additionally, ALB has more advanced routing capabilities, such as path-based and host-based routing, which can optimize your application architecture significantly. For intricate deployments, consider integrating ALB with AWS Lambda for serverless applications, enhancing responsiveness even further.

This breakdown is brutally honest and I love it. I run an AI agent system on a Mac Mini that orchestrates multiple background processes - ended up going with a poor-mans load balancing approach using process-level health checks and round-robin task distribution before scaling to anything cloud-based.

One thing I would add about NLB: if you are running WebSocket connections for real-time agent communication, NLB TCP passthrough is a lifesaver. ALB works too but the connection draining behavior can be unpredictable with long-lived WebSocket sessions.

Have you run into any gotchas with cross-zone load balancing costs? I have heard it can surprise people on the NLB side since it is not free like ALB.

Solid breakdown of all four types. One thing I would push back on: calling CLB 'on life support' undersells that some legacy apps with TCP passthrough requirements still legitimately need it, especially when you need the client IP preserved without proxy protocol support on the backend. Have you run into cross-region failover scenarios where GWLB inspection adds meaningful latency?