Queue depth
Queue depth is the number of outstanding I/O requests at a specified boundary, while a configured depth limit is the maximum or target allowed in flight.
Distinguish in-flight requests from waiting-only counts and queue capacity; report achieved depth rather than treating a configured fio iodepth as an observation. Little's law uses time-averaged occupancy and a matching per-operation mean residence time, not a configured maximum or a tail percentile [4].
Request merging changes the population of operations being timed. Riska et al. measured different request sizes under different schedulers: lower per-request response time did not necessarily mean higher application throughput. Their 2007 SCSI-HDD experiments do not establish an optimal SATA NCQ depth. [5]
MQSim separates host queue capacity, synthetic offered concurrency and per-stream controller fetch admission. Its pinned NVMe implementation can return a fetch credit before a completion is delivered to the host; count and timing boundaries must match when comparing depths. [6] [7]
Relations
- related
- Poisson arrivals see time averages, IOPS, NVMe, Direct I/O, Open-loop load, Closed-loop load, Storage latency, Little's law, I/O request size, Native Command Queuing, NVMe submission queue, NVMe completion queue, I/O scheduler, Sequential I/O, Random I/O, blk-mq request tag, MQSim, Bandwidth-delay product, TCP congestion window, Backpressure, Admission control
Citations
- John D. C. Little. Little's Law as Viewed on Its 50th Anniversary. Operations Research 59(3), 536-549 (2011), pp. 537-539. DOI: 10.1287/opre.1110.0940.
- Riska, Larkby-Lahet and Riedel (USENIX ATC 2007): Evaluating Block-level Optimization through the IO Path, sections 2, 5 and 6
- CMU-SAFARI MQSim README, commit 51f0f2d3fed92d88ef4a0fa61a38024b07bf9d16: device configuration and synthetic workload parameters.
- CMU-SAFARI MQSim Host_Interface_NVMe.cpp, commit 51f0f2d3fed92d88ef4a0fa61a38024b07bf9d16: fetch admission and serviced-request handling.