There is a question almost nobody asks until it is too late: if your primary server vanished right now, how long would it take you to get back to production? Not how long your backups take to complete — you know that — but how much time would pass from the disaster until your users are working normally again. That figure is your real RTO, and it is usually far larger than the one written in the continuity plan.
Having backups is not the same as having recovery capability. A company can complete its copies every night without a single failure for years and still discover on the day of the incident that it needs five days to restore its data, because the entire copy lives on remote storage at the far end of a line that cannot keep up. The backup was perfect; the recovery, unworkable within the window the business could tolerate.
In this article we do the math for real. We work out how long it takes to move 10 TB over lines of different speeds, identify the bottleneck that almost always goes unnoticed, review the factors that stretch real RTO beyond simple transfer, and — most importantly — lay out the strategies that cut that time from days to minutes.
Paper RTO Versus Real RTO
RTO (Recovery Time Objective) is the maximum time your business can tolerate a system being down before the impact becomes unacceptable. It is an objective: the business sets it, not the technology. "Our ERP has a four-hour RTO" means that after four hours of downtime the losses become critical. So far, it is all theory on paper.
Real RTO is a different beast: it is what you actually take, stopwatch in hand, from the moment the disaster is declared until the service is operational and validated again. It includes detecting the incident, deciding to activate the plan, provisioning hardware, locating the right copy, transferring it, rehydrating it, mounting it, booting the systems in the correct order and verifying that everything works. The sum of all those steps almost never fits within the objective someone wrote in a document.
The gap between the two is dangerous because it stays invisible until the day of the incident. If you have never run a full restore and timed it, your "four-hour" RTO is a hypothesis, not a fact. And when the hypothesis fails, it fails at the worst possible moment. Before going further, it helps to be clear on how RTO relates to its sibling, RPO, which we develop in our guide to disaster recovery, RPO and RTO.
The Math of Restoring 10 TB
Let's start with the most common unit mistake: links are measured in bits per second, but data in bytes. A byte is 8 bits, so a 1 Gbps line (1,000 megabits per second) does not move 1,000 but roughly 125 megabytes per second at best. Ignoring that factor of 8 leads to underestimating restore time by almost an order of magnitude.
With that clear, let's do the math for 10 TB of data. A 100 Mbps line delivers at most 12.5 MB/s; a 1 Gbps line, 125 MB/s; a 10 Gbps line, 1,250 MB/s. The theoretical transfer is easy to calculate, but in practice you never reach that maximum: TCP overhead, latency, link contention and the write speed of the destination storage itself cut effective throughput down to 65-80%. The table shows both figures:
| Line | Usable throughput | 10 TB (theoretical) | 10 TB (realistic) |
|---|---|---|---|
| 100 Mbps | ~12.5 MB/s | ~9.3 days | ~12 days |
| 1 Gbps | ~125 MB/s | ~22 hours | ~28 hours |
| 10 Gbps | ~1,250 MB/s | ~2.2 hours | ~3 hours |
The numbers speak for themselves. If your whole 10 TB copy lives behind a 100 Mbps line, your real transfer RTO is almost two weeks, no matter how good your backup software is. And this is only the time to move the bits: we have not yet rehydrated, mounted or booted anything.
The Real Bottleneck: Download Bandwidth
Here is the misunderstanding that costs dearly. Most people size their strategy around the upload (how long copies take to reach the offsite destination), because that is what they see every night. But on the day of the disaster, what matters is the download: pulling the data back from that remote repository into your infrastructure. And many business lines are asymmetric, with far lower upload than download — or the reverse, in the case of dedicated fibres poorly sized for recovery.
The thing almost nobody measures:
Your RTO bottleneck is rarely the speed of the backup. It is the download bandwidth from the offsite or cloud storage to the recovery site. You can copy every night without a hitch and still take days to restore.
Beyond raw bandwidth, a cloud restore brings a second, less obvious bottleneck into play: the provider's own infrastructure can throttle download speed, and in public clouds every gigabyte you pull out is billed as egress. In other words, you pay to recover your own data — and, on top of that, you often do it slowly. Against that model, an offsite backup in a datacenter with a high-capacity network and no egress charges completely changes the arithmetic of recovery.
Factors That Stretch Real RTO
Transfer is only one part. These are the factors that, added together, turn an RTO "of hours" into an RTO "of days":
- check_circle Network and destination storage speed: a 10 Gbps link is useless if the storage you are writing the restore to can't exceed 200 MB/s. The slowest link sets the pace.
- check_circle Rehydration of deduplicated and compressed data: the copy is small because it is deduplicated and compressed, but on restore the full data has to be reconstructed, which consumes CPU and time on top of the raw transfer.
- check_circle Egress and its cost in public clouds: pulling data out of a hyperscaler is slow and billable. The RTO stretches and, on top of that, the recovery bill can come as a surprise.
- check_circle Mount and boot latency: once the data is restored, you have to mount volumes, register the machines with the hypervisor and boot them. That cold start adds minutes or hours depending on the number of systems.
- check_circle Boot order of dependent systems: you can't bring the ERP up before its database, or the application before Active Directory and DNS. The correct order chains waits that add up.
- check_circle Resources available at the recovery site: if you have to source hardware on the spot, that provisioning adds to the RTO. A recovery site that is already prepared removes that delay entirely.
Strategies to Reduce Real RTO
The good news is that real RTO is not a fixed destination: it is designed. The following strategies, combined, let you go from days to minutes depending on how critical each workload is.
- arrow_right Local backup in addition to offsite: keep a nearby copy on the local network to restore quickly from it, and keep the cloud as a last resort against the total loss of the site. The 3-2-1 rule isn't red tape: it is what lets you choose the fastest recovery path available.
- arrow_right Instant Recovery / Instant VM Recovery: instead of waiting for the whole machine to restore, the VM is booted directly from the backup repository and the data is restored in the background while the service is already running. RTO stops depending on the size of the data.
- arrow_right Replication and CDP for the critical stuff: for workloads that cannot tolerate downtime, an always-ready replica or Continuous Data Protection (CDP) with Veeam keeps a near-real-time copy, with an RTO of seconds to minutes.
- arrow_right DRaaS with failover in the provider's datacenter: with Disaster Recovery as a Service you restore nothing: the copy is already in the provider's datacenter and, in a disaster, you boot it there directly. Download bandwidth stops being your bottleneck.
- arrow_right Initial seeding and physical shipment: the first full copy is transferred on a physical disk shipped to the datacenter, avoiding weeks of upload over the network. The same technique works in reverse to speed up a massive restore.
Put into a table, the impact of each strategy on RTO and its relative cost is clear at a glance:
| Method | Typical RTO | Relative cost |
|---|---|---|
| Restore from offsite/cloud (only) | Hours to days | Low (included) |
| Local backup + offsite copy | Minutes to a few hours | Medium |
| Instant Recovery / Instant VM Recovery | Minutes | Medium |
| Replication / CDP | Seconds to minutes | Medium-high |
| DRaaS with failover at the provider | Minutes | Medium-high |
RTO vs RPO and the Importance of Testing the Restore
It is worth not confusing the two metrics that govern any recovery plan. RPO (Recovery Point Objective) measures how much data you can afford to lose — it sets how often you need a copy — while RTO measures how long you can be down. One answers "how much work do I lose?" and the other "how long am I stopped?". Sizing your recovery infrastructure well requires both to be clear; we develop them in detail in the disaster recovery, RPO and RTO guide.
But the single most important point in this whole article is this: an RTO that has not been tested does not exist. The only way to know your real RTO is to run a full end-to-end restore and time it, ideally on a regular, automated basis. Tools like SureBackup in the Veeam Data Platform let you boot copies in an isolated environment and verify they are recoverable without touching production. Testing the restore turns your RTO from a promise into a measured number, and it uncovers bottlenecks while you can still fix them.
EasyDataHost: Fast Recovery With a Site in Spain
At EasyDataHost we design our backup and disaster recovery services to attack real RTO directly, not just the one on paper:
- arrow_right Our DRaaS keeps your workloads ready to boot in our datacenter: in a disaster you fail over in minutes, without restoring 10 TB over the network.
- arrow_right Offsite backup with Veeam Cloud Connect includes Instant Recovery to bring machines up directly from the repository while they restore in the background.
- arrow_right High-capacity network and no egress charges: recovering your data is not billed per gigabyte or throttled, unlike public clouds.
- arrow_right Recovery site in our own datacenter in Spain, with ISO 27001 certification and ENS compliance, low latency to your infrastructure and 24/7 technical support.
If you want to know what your real RTO is and how to reduce it, contact our team: we analyse your data volume, your available bandwidth and your critical workloads to design a recovery strategy sized to the window your business can tolerate.
Frequently Asked Questions
How long does it take to restore 10 TB from a cloud backup?
It depends on your download bandwidth. On a fully saturated 1 Gbps line it is about 22 hours of theoretical transfer (closer to 28 with real overhead); on 100 Mbps the transfer alone exceeds 9 days; on 10 Gbps it drops to roughly 2-3 hours. On top of that, add rehydrating the deduplicated and compressed data, mounting the volumes and booting the systems in the correct order.
What is the difference between paper RTO and real RTO?
Paper RTO is the objective your recovery policy sets. Real RTO is how long you actually take, measured end to end with a timed restore test. The two usually diverge because of the download bandwidth from the offsite, the cost and slowness of egress in public clouds, and the boot order of dependent systems.
How can I dramatically reduce my RTO?
By combining local backup (fast restore from what is nearby), Instant Recovery to boot the VM from the repository itself, replication or CDP for the critical stuff, and DRaaS with failover in the provider's datacenter, where you don't restore: you boot there. And above all, by testing and timing a full restore before you need it.
Conclusion
Having backups is the first step, not the last. What saves a business after a disaster is not the copy, but the speed at which it can get back to production. And that speed is designed in advance:
- arrow_right Real RTO almost always exceeds the paper one; the only way to know it is a timed full restore.
- arrow_right The bottleneck is not the backup: it is the download bandwidth from the offsite. Remember bits vs bytes: 10 TB over 1 Gbps is ~28 real hours.
- arrow_right Rehydration, egress and boot order stretch the RTO well beyond the simple transfer of data.
- arrow_right Instant Recovery, replication/CDP and DRaaS decouple RTO from the size of the data and cut it from days to minutes.