Home / Tools / IOPS & Throughput

IOPS / Throughput Converter

Six quick calculators for the numbers that come up in workload analysis and replication planning.

MB/s ↔ IOPS

Block size (KiB) Throughput (MB/s) — leave empty to convert from IOPS IOPS — leave empty to convert from MB/s

Enter either throughput or IOPS.

Formula: IOPS = MB/s × 1024 / block size (KiB). MB/s = IOPS × block size / 1024.

Workload envelope: throughput + outstanding I/O

Translate an observed or target IOPS rate into throughput, then estimate how many I/O requests must be in flight at the stated average latency.

IOPS Average block size (KiB) Average end-to-end latency (ms)

Enter IOPS, block size, and latency.

Formulas: throughput = IOPS × KiB ÷ 1024; outstanding I/O ≈ IOPS × latency seconds (Little's Law).

Interpret carefully This is a steady-state mathematical envelope, not an ONTAP performance guarantee. Mixed block sizes, burstiness, queueing, cache behavior, protocol overhead, and host-side latency can make real measurements differ. Use the same measurement interval for IOPS and latency.

Mixed workload throughput

Turn a read/write IOPS target into a more realistic payload rate when reads and writes use different average block sizes.

Total IOPS Read mix (%) Average read block size (KiB) Average write block size (KiB)

Enter IOPS, read mix, and block sizes.

Formula: read MiB/s = total IOPS × read fraction × read KiB ÷ 1024; write MiB/s uses the remaining IOPS and write block size.

Payload arithmetic only This does not model WAFL, RAID, network, protocol, compression, deduplication, or storage-system limits. Derive the mix and average block sizes from the same measurement interval, then validate against host and ONTAP counters.

SnapMirror catch-up estimator

How much sustained throughput do you need to move (or catch up) a dataset within a time window?

Data size to transfer (TiB) Time window (hours) Transfer efficiency (1.0 = raw bytes, 1.3–2.0 with compression/dedupe-friendly data)

Enter data size and a time window.

Formula: MB/s = TiB × 1,048,576 / (hours × 3600) / efficiency. This is a planning estimate — real SnapMirror throughput depends on network, source load, and change rate.

Latency budget + concurrency

Test whether a target IOPS rate is mathematically possible with the measured average latency and available outstanding I/O.

Target IOPS Average end-to-end latency (ms) Available outstanding I/O (queue depth × active queues)

Enter target IOPS, latency, and available outstanding I/O.

Formulas: required concurrency = IOPS × latency seconds; mathematical ceiling = outstanding I/O ÷ latency seconds.

Ceiling, not a forecast This applies Little's Law to average values. It does not account for latency percentiles, burstiness, protocol limits, CPU, media, path policy, or ONTAP limits. Treat the result as a consistency check, then validate with host and storage measurements.

Network link headroom

Check whether a measured or planned payload rate fits on a link after reserving room for protocol overhead and traffic bursts.

Payload throughput (MiB/s) Nominal link speed Usable-link assumption (%)

Enter payload, link speed, and a usable-link assumption.

Formula: payload bit rate = MiB/s × 1,048,576 × 8 ÷ 1,000,000,000. Usable capacity = nominal Gb/s × usable-link percentage. This is link arithmetic, not an ONTAP or network performance guarantee.

Choose the percentage deliberately The calculator does not guess protocol overhead. The percentage is your planning assumption for all framing, transport, contention, and burst headroom. Measure the real path before committing to a replication window or workload target.

Reality check A 10 GbE link carries ~1,100–1,200 MB/s of application payload at best. If the estimator demands more than that, you need link aggregation, compression on the relationship, fewer/smaller snapshots, or a longer window.

← All tools