Route 53, CloudFront, and Global Accelerator all sit in front of regional AWS endpoints and steer users to them. This page compares them first, then covers each service.

Choosing a service

Three AWS services put traffic in front of regional endpoints. They differ in what they hand the client and how fast they react to failure.

Route 53CloudFrontGlobal Accelerator
What the client getsA DNS answer chosen by routing policyContent from the nearest edge cacheTwo static anycast IPs
TrafficGoes direct to the chosen endpointHTTP(S); misses go to the originTCP/UDP over the AWS network
Failure handlingHealth checks drop records; clients may cache by TTLOrigin errors surface as 5021Reacts instantly to endpoint health
Best fitName resolution and DNS failoverCacheable web contentLatency- or availability-critical apps, non-HTTP

The Global Accelerator note itself recommends it over DNS-based failover alone for availability-critical apps, and suggests keeping DNS TTLs short in front of it.2 Analysis: the DNS TTL is what limits Route 53 failover, since clients keep a cached answer until it expires; the Route 53 note’s advice to lower TTLs before changes reflects the same limit.3

Amazon Route 53

Route 53 is AWS’s DNS service with three functions: domain registration, DNS routing, and health checking. A query lands in a hosted zone, a routing policy picks which records to answer with, and health checks can remove unhealthy targets before a client sees them.3

Concepts

  • Public and private (VPC-internal) hosted zones.
  • Alias records map a name to an AWS resource such as ELB, CloudFront, or S3, with no charge and no TTL issues; prefer them to CNAME or A records for AWS targets.
  • Routing policies: simple, weighted, latency, failover, geolocation, geoproximity, and multivalue.
  • DNSSEC signs zones against spoofing; VPC Resolver and DNS Firewall handle private resolution and outbound filtering.3

Practices

  • Combine health checks with failover routing across Regions, and alarm on health check status in CloudWatch.
  • Lower TTLs before planned changes.3

Troubleshooting

SymptomCheck
Not resolvingNS delegation at the registrar, records exist, TTL caching
Failover not workingHealth check status, evaluation windows, failover records
Alias errorsCorrect hosted zone ID of the target, fully qualified name
Private DNS fails in VPCVPC DNS attributes, Resolver rules, VPC association
DNSSEC issuesKey signing and DS record at the registrar

As tabled in the note. Default quotas are 500 hosted zones and 10,000 records per zone, adjustable.3 See AWS global traffic routing.

Amazon CloudFront

CloudFront is a content delivery network (CDN). Each request is routed to the lowest-latency edge location; a cache hit never touches the origin, and a miss is fetched from the origin. That makes cache configuration the main lever over cost and freshness.1

Concepts

  • A distribution maps a domain to origins (S3, ELB or API Gateway, custom HTTP servers) and cache behaviors.
  • A cache behavior sets path patterns, TTL (default 24 hours, minimum 0), and which headers and cookies are forwarded.
  • Invalidations remove cached objects before their TTL expires.
  • Two independent gates for private content: Origin Access Control (OAC) stops direct access to the origin, and signed URLs or cookies stop CloudFront serving to unauthorized viewers.1

Practices

  • S3 origin with OAC; signed URLs or cookies rather than public buckets.
  • Set Cache-Control and avoid forwarding headers or cookies you don’t need; a higher hit ratio means fewer origin fetches.
  • ACM certificate with HTTPS enforced; access logs; WAF.1

Troubleshooting

SymptomCheck
Content not updatingTTL and cache behavior; invalidate changed paths
403 from S3 originOAC or OAI and the bucket policy
502 from originOrigin health, custom origin settings, security groups
Slow first byteOrigin latency and hit ratio

As tabled in the note.1 See AWS global traffic routing.

AWS Global Accelerator

Global Accelerator gives an application static anycast IP addresses and routes traffic over the AWS global network to the best regional endpoint by health, client location, and your weights.2

Concepts

  • An accelerator has two static IPv4 addresses (four for dual-stack) for its whole lifetime; deleting it releases them.
  • Listeners (TCP/UDP ports) route to regional endpoint groups of NLBs, ALBs, EC2 instances, or Elastic IPs, weighted and health-checked.
  • Custom routing accelerators map users to specific VPC subnet private IPs, for gaming and real-time apps.
  • It reacts instantly to endpoint health changes.2

Practices

  • Use it for global, latency-sensitive, or availability-critical apps instead of DNS failover alone.
  • Endpoint groups in several Regions with weights for active/passive or active/active.
  • Protect production accelerators from deletion with IAM or tag policies.2

Troubleshooting

SymptomCheck
Clients cannot connectListener ports, endpoint group health
Traffic to an unhealthy RegionEndpoint health and weights
Endpoint unreachableSGs and NACLs allow Global Accelerator’s published ranges
No performance gainDNS points at the static IPs

As tabled in the note.2 See AWS global traffic routing.

Footnotes

  1. Amazon CloudFront - Runbook & Reference, original ↩ ↩2 ↩3 ↩4 ↩5

  2. AWS Global Accelerator - Runbook & Reference, original ↩ ↩2 ↩3 ↩4 ↩5

  3. Amazon Route 53 - Runbook & Reference, original ↩ ↩2 ↩3 ↩4 ↩5