IOPS explained

What affects IOPS?

Several factors affect the IOPS a storage device can provide.

• Block size: Smaller blocks take less time to read or write, so more operations fit into each second, raising IOPS. Larger blocks take longer per operation, which lowers IOPS but raises throughput instead, since each operation moves more data.
• Queue depth: This is the number of I/O requests a device can work on at once. A higher queue depth lets a drive process more requests in parallel, which typically increases sustained IOPS, especially on SSDs and NVMe drives built to handle many simultaneous requests. This holds true up to the device's optimal queue depth. Beyond that point, additional queued requests increase latency without improving throughput and the drive is saturated. For NVMe drives, optimal queue depth can be in the hundreds. For SATA SSDs, it's typically 32 and for HDDs, it's much lower. Monitoring sustained queue depth against the device's rated optimal is one of the clearest ways to confirm a storage bottleneck before latency becomes user-visible.
• CPU capacity: The processor handling read and write requests has to keep up with the storage device. Core count, thread count, clock speed, and on-chip memory all factor in.
• Type of storage device: Mechanical HDDs are limited by platter spin speed and head movement, which typically limits them. SSDs have no moving parts, so they reach much higher numbers.
• RAID level: Depending on the redundancy used in your RAID level, the same data may have to be written more than once, which costs IOPS. This is known as the RAID penalty.

Why does IOPS vary from the provided specification?

Manufacturers usually publish IOPS ratings under ideal lab conditions. Usually under 100% read conditions or a specific 70/30 read/write mix. Write operations cost more than reads on most storage devices, and some RAID levels amplify writes further. When comparing a spec sheet number to a production workload that is write-heavy, expect real-world IOPS to be vary.

The type of I/O operation also matters. Random I/O accesses are scattered across different data blocks. This is typical for databases, OS operations, and transactional workloads. Sequential I/O accesses read or write continuous streams of data that is typical of backups, video streaming, and large file transfers.

SSDs handle both well. HDDs suffer significantly on random I/O because the read head has to physically seek across the platter between operations, which is why HDD IOPS ratings are much lower than their sequential throughput suggests. This distinction is why the same drive can produce very different IOPS numbers depending on the workload pattern.

Interpreting IOPS numbers

A higher IOPS number doesn't automatically mean better performance. Context matters more than the raw figure.

Smaller files are handled efficiently with a smaller block size, since less time is needed to access each block. That setup naturally produces more operations per second, so a high IOPS value matters most here. Larger files are handled more efficiently with a larger block size instead, which means fewer, bigger operations. In that case, high throughput matters more than a high IOPS count.