Q29: Routing a Multi-Region UDP VoIP Service to the Lowest-Latency Region With Failover

Problem Info

Domain	Design Resilient Architectures
Services	AWS Global Accelerator, Network Load Balancer, EC2 Auto Scaling
Difficulty	Medium

Question

A VoIP service uses UDP connections, running on EC2 instances in Auto Scaling groups across multiple AWS Regions. The company needs to route users to the Region with the lowest latency, plus automated failover between Regions.

Which solution meets these requirements?

Correct Answer

Deploy a Network Load Balancer (NLB) with an associated target group tied to the Auto Scaling group in each Region. Register each Region’s NLB as an endpoint behind an AWS Global Accelerator.

Why this is correct

Two requirements point straight at AWS Global Accelerator: routing to the lowest-latency Region, and automated cross-Region failover — both are Global Accelerator’s core features. It uses two static anycast IP addresses that route incoming traffic over AWS’s private global network backbone to the closest healthy endpoint, measured by real network latency, not simple geography. If an endpoint (or an entire Region) becomes unhealthy, Global Accelerator automatically redirects traffic to the next-best healthy endpoint — instant, automated failover with no DNS propagation delay (unlike Route 53-based failover, which is subject to DNS TTL caching on clients).

The UDP requirement is the second key clue: Global Accelerator supports TCP and UDP traffic, unlike CloudFront, which is HTTP/HTTPS (Layer 7) only — ruling out a CloudFront-based design for a VoIP workload. The endpoints behind Global Accelerator need to be load balancers or EC2 instances/Elastic IPs; a Network Load Balancer is the right choice here because NLB operates at Layer 4 and natively supports UDP traffic to the Auto Scaling group’s EC2 targets, whereas an ALB is HTTP/HTTPS-only and wouldn’t support the VoIP service’s UDP protocol at all.

Why the alternatives fall short

• Use Route 53 latency-based routing directly to each Region’s NLB, with health checks for failover — a valid multi-Region latency-routing pattern in general, but DNS-based failover is slower (bound by DNS TTL and client-side caching) compared to Global Accelerator’s near-instant failover at the network layer, which better satisfies “automated failover.”
• Deploy an Application Load Balancer in each Region behind Global Accelerator — ALB is Layer 7 (HTTP/HTTPS) and does not support UDP, making it incompatible with a VoIP service’s UDP-based traffic.
• Use Amazon CloudFront with multiple origins across Regions — CloudFront is built for HTTP/HTTPS content delivery and caching, not for routing arbitrary UDP traffic; it’s the wrong service class entirely for a real-time VoIP workload.

Exam Tip

“UDP” + “multi-Region” + “lowest latency routing” + “automated failover” → AWS Global Accelerator with Network Load Balancer endpoints. Remember the Layer/protocol distinctions: CloudFront = HTTP/HTTPS content delivery (Layer 7, caching). Global Accelerator = TCP/UDP traffic acceleration and routing (Layer 3/4, no caching) — the go-to whenever a non-HTTP protocol like UDP appears alongside multi-Region latency/failover requirements.