Benchmarks

We clocked it. Here's every second, on the record.

A sub-40-second Azure DR failover. Faster incrementals than the enterprise incumbent. Real drills, real numbers, zero spin — see the Azure DR failover results and the server-backup showdown below, then run the same test against your own workload.

Benchmark summary

The three numbers that matter most

First full backup

260.65s

Our one honest loss: the incumbent finished its first full backup faster, at 138s. It's a one-time cost — every backup after it is where XReplicator pulls ahead.

Clean incremental

10.97s

More than 2x faster than the 26s baseline — every incremental after the first counts.

Small change in a large file

17.64 MB

For a 4 GiB file, versus 4.35 GiB shipped by the baseline — that's 99.6% less data over the wire.

Azure DR failover benchmark

From disaster to running VM in under a minute

Azure DR

Attach staged disks as-is

Under 40 seconds

Recovered Azure VM creation after healthy staged DR disks were already synced.

Azure DR

Create disks from snapshots

Around 70 seconds

Recovered Azure VM creation using isolated snapshot-derived disks.

How we ran the drill

No cherry-picking here — this is the exact setup behind the number. Change the workload, readiness state, region, or success criterion and the result will move, so run it your way and see what you get.

Workload
One Azure VM with a 30 GB operating-system disk and a 4 GB data disk.
Readiness
DR sources were healthy and staged disks were synced before failover.
Measured event
Time to create the recovered Azure VM through the selected blueprint path.
Product path
XReplicator v1.3.1 Azure-to-Azure DR orchestration.

What that clock actually measures

The published number is recovered VM creation through the selected XReplicator Azure DR workflow, once staging was ready to go.

What your real-world RTO also needs to cover

Azure control-plane response, capacity, operating-system boot, networking, identity, DNS, application startup, and business validation. Budget for these — we do.

Server backup benchmark

Same VM. Same workload. We put it head-to-head anyway.

Every row below is a side-by-side run against a controlled enterprise backup-tool baseline — CPU is average / maximum, RSS is the maximum observed agent process resident memory, and wire TX is measured on the protected VM. It's a point-in-time lab result, not a claim about every version, configuration, or workload — so judge it on the numbers, not the label.

Fresh full

XReplicator used less average CPU; the baseline completed the first full faster.

Timelower is better
XReplicator260.65s
Enterprise baseline138s
Wire transferlower is better
XReplicator16.23 GiB
Enterprise baseline16.20 GiB
Full measurements
Logical / payload:
XReplicator 25.75 GiB / 16.56 GB; baseline 19.1 GB / 16.1 GB
CPU avg / max:
XReplicator 21.17% / 63.66%; baseline 34.39% / 77.49%
Max RSS:
XReplicator 3.66 GiB; baseline 717 MiB

Clean 1 GiB incremental

XReplicator completed faster and processed only the changed scope in the measured CBT case.

Timelower is better
XReplicator10.97s
Enterprise baseline26s
Wire transferlower is better
XReplicator1.01 GiB
Enterprise baseline1.02 GiB
Full measurements
Logical / payload:
XReplicator 1.50 GiB / 1.025 GB; baseline 17 GB / 1 GB
CPU avg / max:
XReplicator 2.47% / 50.35%; baseline 24.34% / 41.01%
Max RSS:
XReplicator 2.43 GiB; baseline 492 MiB

Small change in a large 4 GiB file

Client-side content deduplication kept existing content off the wire; the restored file checksum matched.

Timelower is better
XReplicator23.87s
Enterprise baseline46s
Wire transferlower is better
XReplicator15.68 MiB
Enterprise baseline4.05 GiB
Full measurements
Logical / payload:
XReplicator 4.01 GiB / 17.64 MB; baseline 4 GiB / 4.35 GiB
CPU avg / max:
XReplicator 25.51% / 31.81%; baseline 25.53% / 43.68%
Max RSS:
XReplicator 2.59 GiB; baseline 683 MiB
Environment: Azure Standard_D4s_v3, Ubuntu 24.04, kernel 6.17.0-1022-azure, 128 GiB protected disk, approximately 16 GiB canonical dataset, and the same backup-server target.
Resource tuning: XReplicator used the reduced profile: batch size 25, one worker, 256 MB configured pipeline budget, and a 120-second dirty-block age gate. Lower batch size is a control knob, not a free performance gain.

Don't take our word for it — run it yourself

Get the complete drill, every variable, and the full validation sequence — everything you need to reproduce this before you set your own internal RTO.

Review full drill