Failover
Health checks and failover
New connections go to healthy endpoints immediately. Existing connections stay on the original endpoint until the idle timeout (TCP 340 seconds, UDP 30 seconds).
AWS Global Accelerator · Network-layer acceleration
Two static IPs put your traffic onto the AWS network at the nearest edge
AWS Global Accelerator places static anycast IP addresses in front of your application. User traffic enters the AWS global network at the nearest AWS edge location and is delivered to healthy Regional endpoints. It works for both TCP and UDP applications.
Standard acceleratorIPv4 · anycast
Network zone 1192.0.2.10
Network zone 2198.51.100.7
Listener
TCP 443client affinity: SOURCE_IPap-northeast-2traffic dial 100%
ALB weight 128 · EC2 weight 128
us-west-2traffic dial 30%
NLB weight 128
IP addresses are examples. The two addresses are served from independent network zones.
2
Static anycast IPv4 addresses assigned per accelerator
99.99%
Monthly uptime SLA for Multi-AZ accelerators
TCP · UDP
Protocols a listener accepts
0–255
Endpoint weight range (default 128)
Why Global Accelerator
Keep the entry point fixed; change paths and endpoints inside AWS
The farther away your users are, the longer the public internet segment and the less consistent the quality. Global Accelerator takes in traffic close to users, moves it onto the AWS network, and keeps the IP addresses clients see unchanged even when you change Regions or endpoints.
Static
IPs that don't change
The IP addresses stay with the accelerator until you delete it. You can change the backend without worrying about firewall allow lists or DNS caching.
Edge
Entry at the nearest edge
Because the addresses are advertised via anycast, traffic is received at the edge location closest to the user and then travels over the AWS global network.
Healthy
Health-based routing
Endpoint health is checked continuously and traffic goes only to healthy endpoints. Unhealthy endpoints are detected and removed from service within one minute.
Traffic control
How traffic reaches an endpoint
A standard accelerator distributes traffic in three tiers: listener, endpoint group, and endpoint. They apply in order from the top.
01Listener
Listener
Accepts incoming connections by protocol and port range. With client affinity enabled, requests from the same source IP go to the same endpoint.
TCPUDPport rangesclient affinity
02Endpoint group
Endpoint group
One per Region. The traffic dial sets the share of traffic that Region receives; setting it to 0% stops traffic to that Region.
1 group per Regiontraffic dial %health check 10s · 30sport override
03Endpoint
Endpoint
Within a group, traffic is split in proportion to weights. ALBs, NLBs, EC2 instances, and Elastic IP addresses can be endpoints.
ALBNLBEC2Elastic IPweight 0–255Features
Key features
Source IP
Client IP address preservation
ALB, EC2, and NLB endpoints with security groups receive the original client IP. It is on by default for ALBs and not available for Elastic IP addresses or NLBs with TLS listeners. Custom routing accelerators always preserve it.
Stateful apps
Client affinity
Connects the same user to the same endpoint based on source IP. Use it for applications that keep session state on the server.
Dual stack
IPv6 support
A dual-stack accelerator gets four static addresses: two IPv4 and two IPv6. Endpoints must be ALBs with client IP preservation enabled, NLBs with security groups, or EC2 instances; Elastic IP addresses are not supported.
Your addresses
BYOIP
Bring your own IPv4 address pool and use it as the accelerator's static IPs, so addresses already on existing allow lists stay the same.
Protection
AWS Shield integration
Because traffic is received at the edge, it works together with AWS Shield DDoS protection. Endpoint security groups and AWS WAF rules remain in place.
Accelerator types
Standard and custom routing accelerators
Choose based on whether AWS distributes the traffic or your application picks the destination itself.
| Type | Destination selection | Endpoints | Health checks · IP | Best fit |
|---|---|---|---|---|
| Standard standard accelerator |
User location, endpoint health, traffic dials and weights | ALB · NLB · EC2 · Elastic IP | Health checks and failover · IPv4 or dual stack | Multi-Region services, blue/green deployments, Regional failure readiness |
| Custom routing custom routing accelerator |
The application uses port mappings to target a specific EC2 instance and port | EC2 instances in VPC subnets | No health checks or failover · IPv4 only · All destinations denied by default | Cases such as game matchmaking or VoIP where users must reach a specific session server |
Reference architecture
From users to Regional endpoints
Users in Seoul and the US connect to the same static IPs, but each enters at a nearby edge location. From the edge, traffic crosses the AWS global network to the per-Region endpoint groups, and traffic dials set each Region's share.
On narrow screens, swipe the diagram sideways.
Getting started
Steps to create an accelerator
- STEP 1
Create the accelerator
Choose standard or custom routing, then IPv4 or dual stack, and BYOIP addresses if needed.
- STEP 2
Add listeners
Set the protocol, port ranges, and client affinity.
- STEP 3
Endpoint groups
Create a group per Region and configure the traffic dial and health checks.
- STEP 4
Register endpoints
Add ALBs, NLBs, EC2 instances, or Elastic IPs and set weights and client IP preservation.
- STEP 5
Switch DNS
Point your application domain at the accelerator's static IPs or DNS name.
Pricing
Cost components
The base costs are a fixed hourly fee per accelerator and a DT-Premium charge on data that crosses the AWS network. Endpoint data transfer is still billed separately.
| Component | Billing basis | Notes |
|---|---|---|
| Fixed accelerator fee | Per accelerator, per full or partial hour | Charged for each provisioned accelerator, whether enabled or not. |
| DT-Premium | Per GB, in the dominant direction only | The rate depends on the combination of the source Region that served the request and the destination edge location. |
| Endpoint data transfer | Standard EC2 data transfer out (DTO) from endpoints | Billed separately from DT-Premium. |
| Public IPv4 addresses | IPv4 addresses used by the accelerator | Standard public IPv4 address pricing applies. |
| Endpoint resources | Usage of each resource such as ALB, NLB, EC2 | Billed at each service's rates. |
Rates vary by Region and usage. See the Global Accelerator pricing page for current rates.
Next step
Design the entry point for your global users
Bring your user distribution, serving Regions, protocols, and failover requirements, and we'll review the accelerator type and traffic distribution design together.