Get a nominal circuit recommendation for disaster recovery, VM replication, database logs, NAS snapshots, object or file sync, and offsite backup. The tool evaluates steady-state and peak replication, RPO-window delivery, initial seeding, and post-outage convergence across workloads sharing one WAN path.
Add servers, VMs, protection groups, or sites. Average and peak use the safe 0–1000%/day range; reduction uses 0–95% or may be left blank to inherit the primary reduction.
Safe range 0–95%; measure locally. Use 0% for unknown, encrypted, or precompressed data.
Safe range 0–100%; includes protocol and product framing.
Safe range 1–100%; lower for loss, latency, retransmits, or product limits.
Safe range 0–100%; AWS recommends at least 20% for burst writes.
| Workload | Average raw | Peak raw | RPO payload | Normal WAN | Peak WAN | RPO |
|---|---|---|---|---|---|---|
| Calculating… | ||||||
Save up to three configurations, such as current link, upgraded link, and peak-change case.
| Scenario | Required circuit | Normal / peak use | RPO feasible | Catch-up time |
|---|---|---|---|---|
| No saved scenarios yet. | ||||
What must be measured: average and peak source write or changed-block rates, realized reduction, end-to-end throughput, and the capacity actually reserved for replication. AWS says required capacity must exceed the combined write throughput of all sources on the same path and recommends collecting at least 24 hours of average and peak measurements using tools such as Windows Performance Monitor or Linux iostat. AWS Elastic Disaster Recovery bandwidth planning
How RPO is modeled: continuous steady-state bandwidth is independent of RPO only under a constant-stream assumption; RPO still limits acceptable lag. Periodic changed-block replication is different: it groups changed blocks by RPO interval, and the bundle must cross the link within its available transfer time. Broadcom also notes that repeated writes to the same block within one interval may transfer that block once, so measured transferred blocks are better than raw I/O when available. Broadcom vSphere Replication guidance
Defaults: the 20% minimum headroom guidance is sourced from AWS; the calculator uses a slightly more conservative 25%. Reduction, overhead, 85% efficiency, 70% WAN share, burst factors, 48-hour offline-seed threshold, and workload presets are planning heuristics. They are deliberately visible and must be replaced with local measurements. Veeam’s report is one example of vendor telemetry that exposes average and aggregate changed data by workload. Veeam ONE Data Change Rate History
Test methodology: deterministic checks cover decimal-unit examples, boundary cases where demand equals capacity, non-converging seed and catch-up cases, multiple-workload aggregation, mode changes, and preset loads. Repository coverage also includes a headless-browser smoke test for default rendering and input-driven recalculation. The model does not simulate vendor scheduling, TCP windows, storage I/O ceilings, journal retention, snapshot stun, or application consistency.
Symbols: D protected bytes, a average daily change fraction, p peak daily change fraction, R RPO seconds, W transfer-window seconds, r reduction fraction, o overhead fraction, e link efficiency, s WAN share, h headroom, L nominal link bps, and B raw backlog bytes.
Cday = D × a; average raw bytes/s is Cavg = Cday ÷ 86,400. Substitute p for peak demand.Crpo = Cavg × R; reduced payload is Prpo = Crpo × (1 − r).Wrpo bits = Prpo × 8 × (1 + o); continuous average wire rate is Cavg × (1 − r) × 8 × (1 + o).scheduled wire bps = Wrpo bits ÷ W. Required nominal WAN is scheduled wire bps ÷ (e × s) × (1 + h).U = L × e × s. RPO is feasible when interval wire traffic transfers within W and average incoming traffic stays below U.B = outage seconds × aggregate average raw bytes/s + queued raw bytes. Required catch-up wire rate is average wire rate + B × reduction/overhead factors × 8 ÷ catch-up seconds.Tcatch = backlog wire bits ÷ (U − average wire bps); it is not finite when new writes consume all usable capacity.Decimal conversion example: 50 GB/hour × 8,000 megabits/GB ÷ 3,600 = 111.11 Mbps before reduction and overhead. Network GB, MB/s, and Mbps are decimal. GiB is binary: 1 GiB = 1,073,741,824 bytes.
Inputs: 10 TB, 5% average and 10% peak daily churn, 35% reduction, 10% overhead, 85% efficiency, 70% WAN share, 25% headroom.
10 TB × 5% = 500 GB/day; average raw = 46.30 Mbps. Wire = 46.30 × 0.65 × 1.10 = 33.10 Mbps. Nominal normal WAN = 33.10 ÷ (0.85 × 0.70) × 1.25 = 69.54 Mbps. The 15-minute raw payload is 5.21 GB; peak nominal WAN is 139.08 Mbps.
Inputs: 50 GB/hour average, 100 GB/hour peak, 20% reduction, 8% overhead, 15-minute RPO, and the same 85%/70%/25% WAN factors.
50 GB/hour = 111.11 Mbps raw. Wire = 96 Mbps; normal nominal WAN = 201.68 Mbps and peak nominal WAN = 403.36 Mbps. Each 15-minute interval contains 12.5 GB raw and 10.8 GB on wire.
Use the VM example with four hours paused and an eight-hour catch-up target. Backlog is 500 GB/day × 4/24 = 83.33 GB raw. Clearing it adds 16.55 Mbps wire to the 33.10 Mbps continuing stream.
Catch-up wire demand is 49.65 Mbps, requiring 104.31 Mbps nominal WAN with the selected factors. A 1 Gbps circuit has ample capacity; a smaller link may meet steady state but miss catch-up.
Collect at least 24 hours of representative average and peak write or changed-block data. Use storage or replication reports, Windows Performance Monitor Disk Write Bytes/sec, Linux iostat, database log-generation reports, or vendor telemetry, then sum every source sharing the WAN path.
With a constant continuous stream, long-term bandwidth follows change rate and RPO mainly limits tolerable lag. Periodic changed-block or scheduled replication packages changes per interval, so RPO and the transfer window determine the burst rate needed to finish each payload on time.
Continuous asynchronous replication sends changes throughout the interval. Periodic snapshots bundle changed blocks and scheduled-window replication must move its interval payload inside a defined transfer window; both can require a higher burst rate than the daily average.
Use average demand to test long-term convergence and a measured peak or high percentile for circuit and RPO-window sizing. Daily averages alone can hide batch jobs, backups, patches, and write bursts.
Mbps is megabits per second and MB/s is megabytes per second. One decimal MB/s equals 8 Mbps. This calculator uses decimal GB and Mbps for network planning and labels binary GiB separately.
Latency, loss, retransmissions, protocol windows, and storage limits can lower realized throughput. Represent this with a conservative link-efficiency value and validate with an end-to-end replication or throughput test.
Use measured reduction from the replication product. Encrypted and precompressed media often reduce poorly, so use zero or a conservative reduction until telemetry proves otherwise.
WAN share limits replication to the capacity actually reserved by QoS or policy. The calculator applies share and efficiency to the nominal circuit before testing workload demand.
Add every server, VM, protection group, or site using the same network path. The calculator applies each row's change rate, RPO, and reduction, then sums normal, peak, and catch-up wire demand.
Consider offline seeding when the full copy does not converge while writes continue, exceeds the operational seed window, or would monopolize the shared WAN for days. Recalculate the reseed after measuring actual reduction and throughput.
Growing lag means average incoming wire traffic exceeds usable replication capacity or the product cannot realize that capacity. Increase reserved bandwidth, reduce change rate, improve reduction, or resolve loss, latency, storage, and software bottlenecks.
Yes. It runs locally in your browser and does not submit inputs to a server.
Use these results for planning and comparison. Validate production replication designs with measured change rate, vendor documentation, packet-loss and latency tests, recovery drills, and business continuity requirements.