← Back to Docs
Recipe

Recipe: Anycast IP routing design

Deploy Meridian nodes behind a shared anycast prefix for zero-failover geographic load balancing without DNS round-robin.

Ingredients

  • Own or leased /24 IPv4 prefix (ARIN/RIPE allocation)
  • BGP peering at 3+ PoPs (Equinix, CoreSite, or cloud on-ramp)
  • Meridian edge nodes at each PoP running identical config
  • Loopback interface bound to the anycast address on every node

Method

  1. Allocate the prefix. Register a /24 with your RIR. This prefix will be announced identically from every edge location.
  2. Configure loopback. Bind the anycast address to lo0 on each Meridian node. Do not assign it to a physical interface — BGP will attract traffic.
  3. Peer with transit. Establish eBGP sessions with upstream providers at each PoP. Announce the /24 with identical attributes (no prepending, same local-pref).
  4. Tune BGP timers. Set hold-time to 6s and keepalive to 2s for sub-10-second convergence on link failure.
  5. Validate with looking glass. Confirm the prefix is reachable via all PoPs. TCP sessions will naturally stick to the closest BGP hop.

Notes

Stateful services require connection tracking sync between PoPs. Meridian's built-in session mirroring handles this automatically when anycast mode is enabled in the node config. Stateless workloads (DNS, HTTP health checks) work out of the box with no additional coordination.

← Return to documentation