AWS Global Accelerator

When you create an accelerator, AWS assigns two static anycast IP addresses from the AWS edge network. Anycast means the same IP address is announced from many edge locations worldwide, so a user is automatically routed to the nearest entry point. These IPs never change for the life of the accelerator, so you can hardcode them in clients, firewalls, or DNS without rotation. You can also bring your own IP address range (BYOIP). The two addresses come from separate network zones, which are isolated units of the AWS network that give you redundancy if one zone has an issue.

The AWS global network is the private, congestion-controlled backbone that connects AWS edge locations and Regions. Once a user request reaches the closest edge location through an anycast IP, Global Accelerator carries it over this backbone to your application endpoint, rather than across the variable public internet. This reduces latency, jitter (variation in delay), and packet loss, and it improves consistency for users who are geographically far from your application Region.

An accelerator contains listeners that process inbound connections on a port or port range using TCP or UDP. Each listener has one or more endpoint groups, and each endpoint group maps to a single AWS Region. Inside a group you register endpoints such as Application Load Balancers, Network Load Balancers, EC2 instances, or Elastic IP addresses. Each endpoint group has a traffic dial (0 to 100 percent) that controls how much of the traffic directed to that Region is allowed through, which is useful for blue/green shifts or draining a Region. Within a group, endpoint weights distribute traffic across individual endpoints.

Global Accelerator routes a request to the optimal endpoint group based on geographic proximity, the endpoint group traffic dial, and endpoint health. By default routing uses a 5-tuple (source IP, source port, destination IP, destination port, protocol) to spread flows. Setting client affinity to source IP keeps all connections from one client going to the same endpoint, which helps stateful applications that need a stable target.

Global Accelerator continuously runs health checks against each endpoint. For ALB and NLB endpoints, it uses the load balancer existing health checks. For EC2 instances and Elastic IPs, you configure the protocol, port, path, interval, and threshold on the endpoint group. When an endpoint or an entire Region becomes unhealthy, Global Accelerator stops sending traffic there and routes to the next-best healthy endpoint group. Because the anycast IPs are static and failover happens at the network layer, recovery is typically within seconds and does not depend on DNS time-to-live (TTL) caching the way a DNS-based failover does.

Both services use the AWS edge network, and they solve different problems. Amazon CloudFront is a content delivery network (CDN) that caches HTTP and HTTPS content at edge locations to serve cacheable web content, media, and APIs close to users. Global Accelerator does not cache; it accelerates and load-balances connections (including non-HTTP TCP and UDP) to your origin over the AWS backbone and gives you static anycast IPs with fast multi-Region failover. Use CloudFront for cacheable web delivery, and Global Accelerator for static IP requirements, non-HTTP protocols, or rapid Regional failover.