Snapshots vs continuous replication: choosing without regrets

TL;DR

Interval snapshots: backup every N minutes, RPO = N. Low cost, RPO from 5 minutes upward. CDP (Continuous Data Protection): writes replicated block by block, RPO < 1 minute. Medium cost. Synchronous replication: every write commits only after confirmation from the remote site, RPO = 0. High cost, low-latency required.

Interval snapshots

How it works

Storage or hypervisor takes a "photo" of the disk at regular intervals (5, 15, 30 min). Deltas are sent to the DR site. Restore starts from the latest consistent snapshot.

Pros

• Simple implementation, universally supported.
• Low bandwidth cost (deltas, not full).
• Compatible with application-aware backup (VSS).

Cons

• Minimum RPO ~5 minutes (below that, frequent-snapshot overhead bites).
• Possible inconsistency across multi-disk apps if not snapshot-aware.

Sweet spot

Systems with RPO > 5 min: file servers, mail (with consistent snapshots), stateless application servers.

Continuous Data Protection (CDP)

How it works

Every write to the primary volume is concurrently sent to the DR site at block or filesystem level. No "snapshots": a continuous stream.

Pros

• 1-30 second RPO at steady state.
• Recovery to an arbitrary point in time (PITR).
• Suited to transactional databases with frequent commits.

Cons

• I/O overhead on the source: typically 5-15% degradation.
• Dedicated bandwidth needed for write-heavy workloads.
• Licence and infrastructure cost: 2-3× interval snapshots.

Sweet spot

OLTP databases, e-commerce with frequent orders, ERPs with continuous writes.

Synchronous replication

How it works

Every write on the primary is confirmed to the application only after the DR site has confirmed receipt. RPO = 0 by construction.

Pros

• Guaranteed zero RPO.
• Transparent failover for many applications.

Cons

• Latency between sites must be < 5 ms (ideally < 2 ms). In Italy feasible between same-datacentre sites or nearby cities with dark fibre.
• Dedicated bandwidth, never the Internet.
• Doubles storage and networking cost.

Sweet spot

Banks, trading, national-security systems. Overkill for 95% of SMBs.

Selection table

| Need | Technology | Relative cost | |---|---|---| | 30-min RPO, file server | Snapshot | 1× | | 5-min RPO, ERP | Frequent snapshot | 1.3× | | 30-sec RPO, database | CDP | 2.5× | | 0 RPO, trading | Synchronous replication | 5-8× |

The Sefthy model

Sefthy uses incremental snapshots with configurable frequency (5 min minimum) for most workloads, and block-level CDP for databases and transactional applications. Synchronous replication is offered on request for specific cases.

FAQ

Can I mix snapshots and CDP in the same system?

Yes. Sefthy supports per-VM policies: CDP only on databases, snapshots for files/apps. Cuts cost 30-40% vs "CDP for everything".

Does CDP impact primary performance?

Yes, 5-15% I/O degradation. For write-heavy workloads (transactional databases), measure first.

Does synchronous replication protect against ransomware?

No. It encrypts the remote site in real time too. For ransomware you need immutable snapshots or air-gap.

To measure real RPO, read RTO vs RPO: differences. For backup vs DR, Backup is not DR.