IOPs

IOPS, or Input/Output Operations Per Second, is a common storage performance metric that measures how many individual read or write operations a storage device or system can complete in one second. IOPS is most meaningful for workloads consisting of small, random I/O requests, such as database operations. It is typically reported as separate read IOPS and write IOPS, and sometimes as a mixed I/O value with a specified read/write ratio.

Measurement and influencing factors: The reported IOPS depends on work load characteristics, including block size, queue depth, and the mix of reads and writes. Synthetic tests often use small block sizes (for example 4 KB) and vary queue depth to simulate concurrent access. Other factors include latency, caching, device type (HDD, SSD, NVMe), interconnects, and RAID configuration. Benchmarking tools such as fio or Iometer are commonly used to generate workloads and measure IOPS.

Interpretation and caveats: IOPS alone do not capture sustained performance or data throughput. A system can deliver high IOPS with tiny data transfers but low bandwidth, or high bandwidth with modest IOPS. Real workloads depend on latency, queue depth, and read/write mix. For capacity planning and performance SLAs, IOPS should be evaluated together with latency, throughput, and the expected workload profile.

Typical ranges: Hard disk drives generally deliver hundreds of IOPS, varying with rotational speed and seek time. SSDs provide thousands to hundreds of thousands of IOPS for small, random operations, and enterprise NVMe SSDs can reach into the low millions under favorable conditions. Actual numbers vary widely with hardware and workload.

Usage: IOPS are used for benchmarking, capacity planning, and performance SLAs, and for comparing storage systems. When evaluating options, tests should reflect realistic workloads and report IOPS alongside latency and throughput.