IOPS Calculator

It starts from the raw backend IOPS your spindles deliver, which is drive count multiplied by IOPS per drive. It then divides that by the average cost per host I/O, where each read counts as 1 backend operation and each write counts as the RAID write penalty. The result is the effective, host-visible IOPS for your read/write mix.

The write penalty is how many backend disk operations a single host write turns into, because parity and mirror RAID levels must update more than one disk. RAID 0 has a penalty of 1 (no redundancy), RAID 1 and RAID 10 are 2, RAID 5 is 4 (read data, read parity, write data, write parity), and RAID 6 is 6 because it carries two parity blocks. Reads always cost 1, regardless of RAID level.

Raw IOPS (also called functional or backend IOPS) is the total the physical drives can sustain, drive count times per-drive IOPS, before any redundancy overhead. Effective IOPS is what your application actually sees after the write penalty is applied to the write portion of the workload. The more writes you have and the higher the RAID penalty, the larger the gap between the two.

Throughput is the effective IOPS multiplied by the I/O block size, then converted to megabytes per second. The calculator uses MB/s = effective IOPS times block size in KiB, divided by 1024. A larger block size moves more data per operation, so the same IOPS yields higher MB/s.

Use the sustained random IOPS of a single drive, not the burst or sequential number. A 7,200 RPM hard drive is roughly 75 to 100 IOPS, a 10K or 15K SAS drive is about 125 to 200, a SATA SSD is in the tens of thousands, and an NVMe SSD can reach hundreds of thousands. Check the vendor datasheet for your specific model and block size.

Reads do not trigger the RAID write penalty, so a workload that is mostly reads has an average operation cost close to 1. A 100% read workload returns effective IOPS equal to the raw IOPS, while a write-heavy workload on RAID 6 can divide your usable IOPS by up to 6. This is why read/write mix matters as much as the drive count when sizing an array.