IOPS vs Throughput
Three Ways to Measure Storage Performance
IOPS vs throughput vs latency compares three different ways to measure storage performance. IOPS counts how many operations complete each second. Throughput measures how much data moves each second. Latency measures how long a single operation takes.
- IOPS counts read or write requests per second. It is covered in more detail in what IOPS is.
- Throughput is the volume of data transferred per second, usually in megabytes or gigabytes.
- Latency is the delay between sending a request and receiving the result, usually measured in milliseconds or microseconds.
A storage system can score well on one metric and poorly on another, so a single number rarely describes it fully.
How the Three Metrics Relate
Throughput equals IOPS multiplied by block size. A drive completing 10,000 operations per second at 4 KB each moves about 40 MB per second. The same drive at 64 KB per operation moves about 640 MB per second. A throughput figure on its own says little about small, random work.
Latency and IOPS are linked through queue depth, the number of requests in flight at once. A drive that handles one request at a time with 0.1 ms latency completes about 10,000 requests per second. If each request takes 1 ms, it completes about 1,000. Queue depth lets a drive work on several requests at once, which raises IOPS without making any single request faster.
Choosing the Right Metric for Each Workload
The workload decides which metric matters most.
- Large file transfers, video streaming and backups depend mainly on throughput.
- Databases, mail servers and sites with many small writes depend mainly on IOPS, because they make many small requests.
- Interactive applications, where each user action waits on a disk read, depend mainly on latency.
Many real systems need all three. Comparing options such as NVMe and SATA SSDs is easier once you know which of the three your application actually stresses.
Frequently asked questions
Which matters more, IOPS or throughput?
Neither matters more in general. Throughput matters when moving large amounts of data in one go. IOPS matters when the workload consists of many small, independent requests. Look at the pattern your application produces, then choose storage based on the metric that matches it.
Is low latency the same as fast storage?
Not entirely. Low latency means each request finishes quickly, which users feel as responsiveness. A drive can also deliver large transfers at high throughput while having higher latency for individual requests. Consider all three metrics together before drawing conclusions about a storage system.