What does this calculator do?
IOPS counts input/output operations per second. This calculator offers the three ways the number is normally derived. From an operation count and a duration it is an exact measurement of what happened. From average latency it is an idealised estimate — 1000 ÷ latency in milliseconds — which describes one operation at a time with no queuing, so it understates what a device with parallelism can do. From a read/write mix it computes backend IOPS: every write costs the array several physical operations, so the disks behind a RAID set must sustain more IOPS than the application asks for. IOPS figures are only comparable when the block size and the read/write mix match; this tool does not derive IOPS from spindle RPM.
Formula
Three modes, three formulas.
IOPS = operations ÷ seconds IOPS ≈ 1000 ÷ latencyMs (idealised, one operation at a time) reads = totalIops × readPct ÷ 100 writes = totalIops − reads backendIops = reads + writes × writePenalty
How to use it
Pick the mode and the form shows only the fields it needs. For the latency mode, use the average service time of a single operation, and treat the result as a ceiling for a serial workload rather than a device rating. For the workload mode, enter the IOPS your application requires, the read share as a percentage, and the write penalty of the RAID level: 2 for RAID 1 and RAID 10, 4 for RAID 5, 6 for RAID 6, 1 if there is no penalty.
Worked example
Operations mode: 120,000 operations in 60 seconds = 2,000 IOPS. Latency mode: 0.5 ms per operation gives 1000 ÷ 0.5 = 2,000 IOPS for a single serial stream. Workload mode: 5,000 front-end IOPS at 70% reads means 3,500 reads and 1,500 writes; on RAID 5 with a write penalty of 4 the backend must sustain 3,500 + 1,500 × 4 = 9,500 IOPS.
Frequently asked questions
Why does the write penalty exist?
Parity RAID has to keep parity consistent. A small RAID 5 write reads the old data and the old parity, then writes the new data and the new parity — four operations for one logical write. RAID 6 has two parity blocks, so the cost is six. Mirroring simply writes twice.
Is the latency mode a device rating?
No. It assumes strictly one operation in flight at a time. Real devices overlap many operations, so an SSD with 0.5 ms latency can deliver far more than 2,000 IOPS at depth. Use the latency figure for serial, latency-bound workloads.
Do IOPS numbers transfer between block sizes?
Not directly. 10,000 IOPS at 4 KiB and 10,000 IOPS at 64 KiB are sixteen times apart in throughput, and most devices cannot sustain the same IOPS at the larger size. Always state the block size with the IOPS figure, and use the Disk Throughput Calculator to convert.
Sources
- SNIA Dictionary — vendor-neutral definitions of RAID levels, stripe, mirror, parity, IOPS and raw versus usable capacity.
- NIST — Prefixes for binary multiples (IEC 80000-13): a drive sold as 2 TB (2 × 10¹² bytes) is about 1.82 TiB as reported by most operating systems, which is why raw and usable figures are shown in the unit you choose.
- Nominal equal-drive capacity model used here: RAID 0 = N × S, RAID 1 = S, RAID 5 = (N−1) × S, RAID 6 = (N−2) × S, RAID 10 = (N/2) × S. Filesystem overhead, metadata, hot spares, vendor reservations, compression and deduplication are not modelled unless you enter them.
- Reviewed: 2026-08-10
Related calculators
Last updated: 2026-08-10 · v1.0.0 · Report an error