Estimate how much storage to provision from measured ingest or current data, retention, compression, copy count, compound growth, snapshots, filesystem overhead, and operating headroom. Use it for logs, databases, backups, analytics, file servers, and object storage.
Estimate drive requirements from application-usable capacity. Layout math is simplified and should be checked against the controller or storage vendor.
Complete the assumptions and calculate.
| Capacity stage | Amount |
|---|---|
| Retained logical source data | — |
| Compression / deduplication savings | — |
| Compressed footprint | — |
| Protected footprint (copies) | — |
| Snapshot allowance | — |
| Filesystem / metadata overhead | — |
| Modeled consumption before reserve | — |
| Free-space reserve | — |
| Required provisioned capacity | — |
| Effective total multiplier | — |
retained source ÷ compression × copies → protected footprint; then add snapshots and filesystem overhead; provisioned capacity = modeled consumption ÷ (1 − free-space reserve)
| Milestone | Required provisioned | Headroom vs installed |
|---|
Complete the assumptions and calculate.
| Installed raw capacity | — |
| Layout-usable capacity before reserve | — |
| Usable after free-space reserve | — |
| Required raw capacity at minimum drive count | — |
| Planned drive surplus / shortfall | — |
Raw means all labeled drive capacity. Layout-usable subtracts parity, mirrors, and hot spares. Reserve is intentionally unused space and is applied afterward.
Edit the uncertain assumptions independently. All other workload inputs—including copies, snapshots, filesystem overhead, and the selected horizon—stay the same. For current-data mode, retention does not affect the calculation.
| Scenario | Growth (%) | Compression (:1) | Retention | Reserve (%) | Required capacity |
|---|---|---|---|---|---|
| Conservative | — | ||||
| Expected | — | ||||
| High growth | — |
ingest per day × retention days. For current-data mode, it is the current logical dataset. Compound growth applies to ingest or current logical data before later stages.retained source ÷ compression ratio. A 1:1 ratio makes no reduction.compressed footprint × copy count. Copy count describes stored copies, not array raw capacity and not independent backup.protected footprint × snapshot %. It is separate because change rate and snapshot retention drive this cost.(protected footprint + snapshots) × filesystem %. This models metadata and allocation costs separately from snapshots.(protected + snapshots + filesystem overhead) ÷ (1 − reserve). The difference is intentionally unused free space.Unit discipline: GB, TB, and PB use powers of 1,000 bytes. GiB, TiB, and PiB use powers of 1,024 bytes. The calculator converts through bytes and displays both systems; it never divides GB by 1,024 and labels the result TB.
Each row starts from the 30-day log example and changes only the named assumption.
| Change | Required provisioned | Difference |
|---|
Text, logs, and repetitive columnar data may reduce substantially. JPEG, video, audio, encrypted data, and ZIP-style archives often do not. Use 1:1 until a representative sample test proves otherwise; Microsoft likewise identifies common archive and media formats as non-compressible for SMB compression.
Test 10–20% as a planning range, then follow platform limits. Elasticsearch's default low and high disk watermarks correspond to 15% and 10% free, while Ceph warns at 85% used and restricts backfill at 90% used by default. Distributed imbalance or rebuild space can require more.
A snapshot allowance cannot be inferred from count alone. Incremental snapshots retain new and changed blocks, so change rate and retention matter. Snapshot copies may also share the source system's failure domain.
Replication and RAID keep services available through selected hardware failures. A separately retained, recoverable backup addresses different events. Do not combine replica count and backup retention unless every copy truly belongs in the same storage tier being sized.
Workload-specific metadata, erasure-coding stripe rules, multiple RAID groups, rebuild workspace, backup-job overlap, thin-provisioning risk, compression block effects, drive failure domains, controller formatting, and vendor-reserved capacity. Add these from platform documentation before purchasing.
Raw capacity is the sum of the manufacturers' labeled drive capacities. Usable capacity is what remains after RAID, mirroring, erasure coding, spares, formatting, and vendor reserves. This workload mode reports required provisioned capacity rather than calling protected data raw or usable.
For an ingest-and-retention workload, growth applies to the ingest rate before multiplying by retention. For a current-data workload, it applies to the current logical dataset. Use the mode that matches the measurement you actually have.
Snapshots preserve point-in-time state and often share the same system and failure domain as the source. Backups are separate recoverable copies with their own retention and failure-domain design. Snapshot capacity depends mainly on change rate and retention.
No. RAID and replication improve availability for selected failures, but deletion, corruption, ransomware, or an administrative mistake can be copied to every replica. Model independently retained backups separately.
Use a ratio measured from representative production data. Use 1:1 when evidence is unavailable or for encrypted, pre-compressed, image, audio, video, and archive data that may not shrink materially.
A decimal TB is 1,000,000,000,000 bytes, while a binary TiB is 1,099,511,627,776 bytes. They describe the same bytes with different divisors, so one TB is about 0.91 TiB.
It depends on the platform and workload. A 10 to 20 percent planning reserve is a useful test range, not a universal rule; rebuilds, rebalancing, write bursts, and vendor thresholds may require more. Follow the most restrictive platform guidance.
Run one calculation per tier with its own ingest, retention, reduction, copies, growth, and reserve. Do not average assumptions across tiers; add their final provisioned requirements only after each tier is modeled.
Yes. Enter current installed capacity in workload mode. The forecast compares the same provisioned-capacity requirement against it and reports the first projected month that demand exceeds capacity, when the selected growth rate reaches it.
Method: transparent logical-data → reduction → protection → snapshots → filesystem overhead → reserve model, with compound source growth.
Technical review: Starlight Robotics Infrastructure Team, maintainers of infrastructure planning calculators.
Last reviewed: 15 July 2026.
Use: suitable for early sizing and scenario comparison. Validate production purchases against measured workload data, failure-domain design, and vendor sizing tools.