AWS Global Accelerator handles TCP termination at the edge by establishing two separate TCP connections to optimize performance and reduce latency. When a client initiates a TCP connection, instead of the connection being directly established between the client and the application endpoint in the AWS Region, Global Accelerator terminates the TCP connection at the AWS edge location closest to the client. This means the initial three-way TCP handshake occurs between the client and the nearest AWS edge location, significantly reducing the connection setup time, especially for clients far from the application endpoint[2][3][4][5][8].
Almost simultaneously, Global Accelerator establishes a second TCP connection from the AWS edge location to the application endpoint in the AWS Region. This upstream connection benefits from running over the AWS global network, which is a well-monitored, congestion-free, and redundant infrastructure optimized for low latency and high throughput. This design allows faster client responses from the edge location while ensuring efficient and reliable data transfer between the edge and the application endpoint[3][4][5].
Additional optimizations that enhance TCP termination at the edge include:
- Jumbo frame support between the edge location and the application endpoint, allowing up to six times more payload per packet, which reduces the total transmission time[5].
- Large receive side window and TCP buffers on the AWS edge infrastructure, enabling the edge to buffer and process larger amounts of data quickly, improving download speeds to clients[5].
- Large TCP congestion window scaling over the AWS global network, which allows sending larger amounts of data than typical internet paths, further boosting throughput[5].
TCP termination at the edge is enabled by default on all Global Accelerators and supports both IPv4 and IPv6 traffic, including dual-stack accelerators, at no additional cost[2][4].
In summary, by splitting the TCP connection into two segments-client to edge, and edge to endpoint-AWS Global Accelerator reduces initial connection latency, improves throughput, and leverages the AWS global network for optimal routing and performance. This results in up to a 60% improvement in throughput and faster response times for internet clients accessing AWS-hosted applications[5][7].
Citations:
[1] https://aws.amazon.com/about-aws/whats-new/2020/03/aws-global-accelerator-launches-tcp-termination-at-the-edge/
[2] https://aws.amazon.com/about-aws/whats-new/2023/05/aws-global-accelerator-extends-tcp-termination-ipv6-traffic/
[3] https://docs.aws.amazon.com/global-accelerator/latest/dg/introduction-how-it-works.html
[4] https://aws.amazon.com/global-accelerator/features/
[5] https://aws.amazon.com/blogs/networking-and-content-delivery/achieve-up-to-60-better-performance-for-internet-traffic-with-aws-global-accelerator/
[6] https://tutorialsdojo.com/aws-global-accelerator/
[7] https://hands-on.cloud/aws-services/global-accelerator/
[8] https://www.projectpro.io/recipes/explain-features-of-aws-global-accelerator
[9] https://en.wikipedia.org/wiki/Transmission_Control_Protocol
[10] https://www.hava.io/blog/what-is-aws-global-accelerator