queuedepth

QueueDepth, in computing, refers to the maximum number of outstanding I/O requests or commands that a storage host, bus, or device queue can hold at once. It defines how many operations can be in flight concurrently and is a key tuning parameter in storage performance.

In most storage stacks, I/O requests from applications enter a host-side queue and are issued to the storage device, which also maintains its own hardware queue. The effective queue depth is influenced by both the host queue and the device queue, as well as the I/O scheduler in use. Modern devices such as NVMe drives expose large, configurable queue depths and multiple hardware queues, while traditional spinning disks and older interfaces have smaller limits.

Impact: Increasing queue depth generally improves throughput by better hiding device latency, up to the point of saturation, after which additional depth yields diminishing returns and can increase latency or CPU overhead. Workload characteristics (random vs sequential, read vs write, queue depth sensitivity) and hardware (SSD vs HDD, number of channels) determine the optimal value.

Configuration and measurement: Queue depth is commonly tuned per device or path in enterprise systems. Practitioners vary depth values during benchmarking to map performance curves. Tools such as fio, iostat, and perf are used to measure throughput and latency as depth is varied. Values often range from single digits to hundreds for SSDs and can reach thousands for high-end NVMe arrays, but benchmarks and vendor guidance should be consulted for specific platforms.

See also: I/O scheduler, queue depth (storage), NVMe, SCSI, iostat, fio.