Rotational latency
Rotational latency is the wait for a requested sector to pass under a hard-disk head after track positioning, modeled as half a revolution on average under uniform phase.
Model: with uniform target-relative phase after positioning, mean wait is half a revolution: 30,000 / RPM ms. This gives about 4.17 ms at 7,200 RPM, 3 ms at 10,000 RPM and 2 ms at 15,000 RPM [2]. A 15,000-RPM product manual specifies 2 ms [3].
Request cost: seeking, rotation and transfer contribute separately. A historical textbook example adds 4 ms seek, 2 ms rotation and about 30 microseconds transfer: roughly 6 ms. Long sequential transfers amortize positioning [2]; queuing and controller work also affect response time.
Specification: a 7,200-RPM SAS manual prints 4.16 ms [4]. Preserve that value separately from calculated 4.17 ms. Neither is complete request latency or a predictor of general random IOPS.
Scope: the platter rotates during seeking. One revolution bounds a single uninterrupted wait for the next sector passage, not total request latency; the uniform-phase maximum approaches 8.33 ms at 7,200 RPM. The reciprocal half-turn mean is a continuously busy serial-model rate, not a workload-independent IOPS ceiling. Locality, read-ahead and rotation-aware scheduling change its assumptions; cache hits may avoid media access [2] [5].
Rate denominator: with no overlap or idle gaps, two chosen waits of 1 ms and 9 ms give 2 completions / 10 ms = 200 operations/s. Averaging their reciprocals gives about 555.6 operations/s, not this throughput. These are constructed values, not HDD measurements. Include idle gaps in elapsed time; overlapping latencies are not additive elapsed time. The linked calculation retains six RPM rows and reproduces this example [6].
Read-on-arrival can collect a full physical track in one revolution after positioning; this is not zero request latency. [7]
Relations
- related
- Hard disk drive (HDD), Seek time, Rotational media, Mission-critical HDD, Nearline SAS, Nearline HDD, Queue depth, Drum memory, Freeblock scheduling, Read-on-arrival, Rotational replication
Advantages
- Spindle specifications support a simple mechanical-delay estimate with explicit assumptions. [2, 3]
- Rotation-aware scheduling can compare candidate requests by positioning cost, rather than seek distance alone. [2]
Limitations
- A mean conceals phase-dependent variation and is not a percentile or per-request bound. [2]
- Increasing RPM does not remove actuator motion or guarantee a proportional improvement in application performance. [2]
Citations
- Arpaci-Dusseau and Arpaci-Dusseau, Operating Systems: Three Easy Pieces, chapter 37: Hard Disk Drives (2008-23)
- Manufacturer product manual, 2.5-inch 15,000-rpm 12 Gbit/s SAS enterprise HDD (Rev. C, September 2020), average rotational latency
- Seagate Exos X18 SAS Product Manual 100865853, Rev. F (May 2021), sections 4.1-4.1.3
- Ruemmler and Wilkes: An Introduction to Disk Drive Modeling, IEEE Computer 27(3):17-29 (March 1994), controller and rotational-position modeling sections
- Hesela: HDD rotational latency by spindle speed, conditional model and unchanged arithmetic rows
- Griffin, Schindler, Schlosser, Bucy and Ganger: Timing-accurate Storage Emulation, FAST 2002, section 6.1 and Table 5
Further reading
Sources & caveats
- The textbook's 4 ms seek example describes a historical drive, not a current nearline measurement [2].
- The 4.16 ms example is Seagate Exos X18 SAS manual 100865853 Rev. F (May 2021), sections 4.1-4.1.3 [4].
- The half-revolution formula is a uniform-phase model; spindle speed alone does not determine every observed request latency.
- The 1994 disk-modeling paper supports phase, scheduling and read-ahead mechanisms, not present-day drive performance.