IOPSes

IOPS, short for input/output operations per second, is a common unit of performance used to quantify how many discrete read and write operations a storage device or subsystem can complete in one second. It is not a measure of data transfer rate; throughput, measured in MB/s, represents the amount of data moved per second, while IOPS focuses on operation count. An individual operation can be a read or a write, and workloads may be random or sequential, with different block sizes.

IOPS are influenced by several factors: the type of storage media (hard disk drives generally deliver lower IOPS than solid-state drives and NVMe devices); the mix of reads vs writes; the block size of operations; the depth of the I/O queue (queue depth); latency; caching and read-ahead; and software/hardware layers such as RAID, virtualization, or storage controllers.

Measurement is typically performed with benchmarks that simulate defined workloads. IOPS are reported under a specified block size and access pattern, such as random 4K or 8K reads/writes. Tools commonly used include fio, Iometer, and vdbench. The measured IOPS can vary widely with workload characteristics, especially random versus sequential access.

Practical use: when designing or evaluating storage, IOPS should be considered alongside latency and throughput. Plans often specify target IOPS at a given read/write mix and block size, while ensuring latency remains within acceptable bounds. Real-world performance depends on concurrent users, queuing, and caching, so provisioned IOPS (or IOPS limits) may be higher than sustained observed values.

Limitations: IOPS is a simplified metric that does not capture latency distribution, tail latency, or the impact of sporadic I/O bursts. It should be used with other metrics to assess overall storage performance.