• All other factors being equal, a lower TTL is preferred
    • This minimizes time to re-direct traffic away from a failed site
    • Should always be at least long enough that DNS operations don’t become a hindrance to application/service access
  • For active/standby site configurations
    • No special considerations
  • For multi-active-site GLB configurations
    • Persistence should be enabled
      • Timer has to be higher than the TTL of the DNS record itself.
        • Otherwise, persistence could expire before it actually gets invoked.
  • If the GLB’s target-pool is populated with FQDNs (and not IP addresses), the TTL value used in the GLB DNS responses to clients must not exceed that of the A records for the target FQDNs.
    • If, for example, a 3rd-party dGLB is configured with an FQDN that load-balances to multiple AWS application-load-balancers
      • AWS DNS always responds to resolution requests for ALB’s FQDNs with a TTL of 60 seconds.
      • If the dGLB is configured with a 30-second persistence timer, the clients (recursive resolvers) would keep the A records cached past the 30 seconds, and the GLB would expire the persistence-able entry before it received another query