Everything worth knowing, nothing that isn't.
What triggers a failover
Three classes of trigger. Hard failures: TCP disconnect to the destination SMSC, SMPP bind expiration, persistent submit_sm error codes (ESME_RSYSERR, ESME_RTHROTTLED). Soft failures: latency p95 exceeding a threshold, DLR success rate dropping below a per-route minimum (e.g. <85%), per-operator congestion windows tightening. Predictive triggers: scheduled maintenance windows on partner routes, known operator-side incidents from upstream signal.
How fast it has to be
For an OTP, sub-second failover. For a marketing campaign, can be 5-30 seconds. BroadNet's routing engine evaluates per-message in under 5ms, applying the cheapest-delivering rule with per-operator quality scores updated in real time. If the primary route is degraded, the secondary is selected before the message ever hits the wire.
The traps
Naive failover routes to the same operator's SMSC via a different aggregator — which is meaningless if the issue is operator-side. Good failover routes to a structurally different path (different SS7 PoP, different transit aggregator, different MNC if the destination supports it). The best failover knows the failure mode and picks accordingly: a TCP-level failure means try the same operator via a different transit; an HLR-level failure means try a different operator-portability path.
What buyers should ask
Ask your aggregator: "Show me your route map and your route-quality scoreboard for the last 30 days, per destination operator." An aggregator that can't show this doesn't have real failover — they have a single route and they pray. BroadNet's customer-facing route dashboard updates every minute. Available to every account, not just enterprise.