Disk IOPS Calculator for RAID, SSD and Workload Throughput
Estimate storage performance from drive IOPS, RAID layout, read/write mix, block size, and controller efficiency. The tool separates small random I/O from sequential throughput so you can plan HDD, SSD, and NVMe arrays without sending inputs anywhere.
Inputs
Results
Core random I/O model: read ceiling = drives x read IOPS x efficiency,
write ceiling = drives x write IOPS / RAID write penalty x efficiency, and
mixed IOPS = 1 / (read fraction / read ceiling + write fraction / write ceiling).
Advertisement
How to use this disk IOPS calculator
- Start with measured drive data. If you do not have measurements, choose a conservative preset for the media type.
- Choose the RAID level. Parity RAID protects capacity efficiently but small random writes pay a write penalty.
- Match the workload. The read/write mix and block size determine whether the array is IOPS-bound or throughput-bound.
- Use efficiency as a reality factor. Controllers, filesystems, snapshots, encryption, queue depth, and latency targets reduce usable performance.
- Benchmark before buying. Final sizing should be validated with realistic queue depth, data reduction, cache state, failure mode, and rebuild tests.
Methodology and assumptions
This calculator uses the common small-block random I/O planning model for RAID write penalties. Random reads are treated as one physical read per logical read and can usually be served by any array member. Random writes on parity RAID require extra reads and writes to update parity, so RAID 5 uses a write penalty of 4 and RAID 6 uses a write penalty of 6. RAID 10 uses a penalty of 2 because each logical write is mirrored.
Sequential throughput is estimated separately because full-stripe writes, caching, request size, and controller behavior can make sequential performance very different from random IOPS. The MB/s values are decimal megabytes per second, while workload data rate from block size is shown in MiB/s using binary KiB and MiB units.
RAID write penalty reference
| Layout | Random read model | Small random write penalty | Planning note |
|---|---|---|---|
| RAID 0 | All drives can serve reads | 1 | No redundancy. Do not use for important data without another protection layer. |
| RAID 1 | Mirrors can serve reads | Mirror count | A two-drive mirror has about one drive of random write IOPS. |
| RAID 5 | All drives can serve reads | 4 | Efficient capacity, weaker for small random writes and large rebuild-risk pools. |
| RAID 6 | All drives can serve reads | 6 | Two-parity protection, but a larger random-write penalty. |
| RAID 10 | All drives can serve reads | 2 | Often preferred for write-heavy databases and VM stores. |
| RAID 50 | All drives can serve reads | 4 | Stripe across RAID 5 groups. Each group needs at least three drives. |
| RAID 60 | All drives can serve reads | 6 | Stripe across RAID 6 groups. Each group needs at least four drives. |
FAQs
Why does the mixed IOPS result differ from read IOPS and write IOPS?
The mixed result applies the workload read/write ratio. A 70/30 workload consumes some read capacity and some write capacity, so the lower side can become the bottleneck.
Why are SSD preset numbers lower than some datasheets?
The presets are intentionally conservative. Many datasheet numbers assume high queue depth, ideal block sizes, empty drives, or cache-friendly workloads.
Does cache change RAID write penalty?
Yes, cache and write coalescing can reduce observed penalties for bursts and full-stripe writes. For sustained small random writes, the penalty model is still useful for planning.
How should I model degraded mode?
Use a lower efficiency value and validate with the storage platform. Rebuilds, parity reconstruction, and spare activation can reduce both IOPS and latency consistency.
Is RAID a backup?
No. RAID improves availability after some disk failures, but it does not protect against deletion, corruption, ransomware, theft, or site loss.
Is this calculator private?
Yes. Calculations run in your browser and inputs are not submitted to a server.
Disclaimer
Use these results for sizing discussions and comparison, not as a production guarantee. Validate final storage designs against vendor documentation, benchmark data, application latency targets, cache settings, endurance limits, and backup or recovery requirements.