IOPS and Throughput Calculator

IOPS is how many read or write operations storage handles per second. Block size is how much data each operation moves. Multiply them and you get throughput:

Throughput (MiB/s) = IOPS x block size (KiB) / 1,024

Bandwidth is the same number in bits (x 8), which is handy when checking whether a network link like iSCSI or NFS can keep up.

Required concurrency uses Little's Law: outstanding I/O = IOPS x latency (in seconds). To get 10,000 IOPS at 1 ms, about 10 requests have to be in flight at all times. If your app only issues one at a time, it will never go faster than 1 / latency.

A database doing 10,000 IOPS at a 16 KiB block size and 1 ms average latency:

• Throughput: 10,000 x 16 / 1,024 = 156.25 MiB/s (about 164 MB/s)
• Bandwidth: 156.25 x 8 = 1,250 Mib/s, more than a 1 Gb link can carry
• Concurrency: 10,000 x 0.001 = 10 outstanding I/O

So this workload needs a 10 Gb storage network or faster, and a queue depth of at least 10.

Spec sheets usually quote 4 KiB random I/O at a deep queue. Bigger blocks, a shallow queue, a write-heavy mix, or a busy controller all lower real IOPS.

Reads are spread across disks, but each write costs extra back-end I/O: about 2 for RAID 10, 4 for RAID 5 and 6 for RAID 6. A write-heavy load on RAID 6 needs far more raw disk IOPS than the front-end number.

The calculator uses binary units (1 MiB = 1,048,576 bytes), like most operating systems. Drive and network vendors use decimal MB, which reads about 5% higher for the same speed.