4.8.1. DR Site Types and Replication Strategies
💡 First Principle: The speed and cost of recovery from a site-level disaster depends entirely on how much preparation you've done beforehand. Hot sites cost the most but recover instantly; cold sites cost the least but take the longest. Your RTO drives which site type is appropriate.
DR Site Types
| Site Type | What's There | Recovery Time | Cost | Notes |
|---|---|---|---|---|
| Hot site | Fully operational duplicate: running hardware, replicated data, live systems | Minutes to hours | Highest | Immediate failover; essentially a second production environment |
| Warm site | Hardware installed and configured, but data must be restored | Hours to days | Moderate | Balance of cost and recovery speed |
| Cold site | Facility with power and cooling; no hardware | Days to weeks | Lowest | Longest recovery; must procure/install hardware |
| Cloud DR | Cloud-hosted replicas that can be activated on demand | Variable (minutes to hours) | Pay-as-you-go | Increasingly common; no hardware maintenance |
Geographic Separation
DR sites should be in a separate geographic location to survive regional disasters (hurricanes, earthquakes, floods). A DR site 10 miles away may be affected by the same event.
Replication Strategies
Replication keeps DR site data current. The replication method determines RPO:
| Method | How It Works | RPO | Use Case |
|---|---|---|---|
| Synchronous | Write confirmed only after both primary and DR sites write | Near-zero (RPO ≈ 0) | Mission-critical: banking transactions, real-time systems |
| Asynchronous | Write confirmed at primary; DR site catches up slightly behind | Minutes to hours | Most enterprise DR; good balance of performance and data protection |
| Application-consistent | Snapshot coordinates with application to ensure consistent state | Snapshot interval | Database servers; ensures DB is in consistent state at recovery point |
| Mirroring | Real-time copy maintained | Near-zero | Critical data sets; requires high bandwidth |
| Bidirectional | Both sites can be active and write; replication flows both ways | Near-zero | Active-active DR; most complex |
| File locking | Prevents simultaneous writes to same file from both sites | N/A | Prevents split-brain in DR scenarios |
⚠️ Exam Trap: Synchronous replication requires the primary write to wait for the DR site to confirm before completing. This adds latency proportional to the network distance between sites. Over a wide geographic distance, synchronous replication can significantly impact application performance.
Reflection Question: A financial services company requires an RPO of less than 5 minutes for its transaction database and has DR sites 1,000 miles apart. What replication method is appropriate, and what performance consideration must be evaluated?
