Dual-Actuator Hard Disk Drive

Host presentation is model-specific. Seagate's FAQ describes Exos 2X SAS as two LUNs, each associated with one actuator. Exos 2X SATA instead exposes one logical device. Its manual maps the first and second halves of the LBA space to different actuators; the Concurrent Positioning Ranges log, address 47h, reports those ranges. This is not an implicit RAID array.

The Linux 6.14 ABI exposes advertised parallel ranges under /sys/block/<disk>/queue/independent_access_ranges/. Each numbered directory contains sector and nr_sectors. Both use 512-byte units regardless of the device's logical block size. Missing metadata is not proof that a physical drive has only one actuator.

For a reported start s and length n, the byte interval is [s * 512, (s + n) * 512), with the upper endpoint excluded. An illustrative start of 2048 and length of 4096 therefore means bytes [1,048,576, 3,145,728), even on a 4Kn device. These invented values demonstrate units, not a measured drive layout.

Dual-actuator is not a synonym for dual-stage. A dual-stage servo combines a coarse voice-coil drive with a fine-positioning actuator near the head; Toshiba's 2011 prototype uses piezoelectric elements on the suspension. Those stages cooperate on positioning, rather than exposing two independently addressable data regions. One HDD can combine both ideas.

Independent positioning also does not mean mechanical isolation. Chen, Prakash and Horowitz (2021, sections I-IV) study a seek on one assembly disturbing another that follows a track through their shared pivot. They numerically evaluate adaptive feedforward compensation and microactuator stroke limits. Their position-error results are not an application-throughput benchmark or a field failure-rate measurement.

Earlier research used a different meaning of two arms. Chandy's 2003 logging-disk simulation assigns one arm to reads and the other to writes, and explicitly assumes only one arm seeks and accesses data at a time. Its low-rate synthetic workload and historical drive models are not evidence of a MACH.2 speedup. Read/write specialization must not be confused with the actuator-specific LBA regions below.

For an engineering evaluation, record the actual address mapping, physical-drive identity, kernel, controller and firmware. Compare concurrent traffic to both regions against a single-region baseline with the same total queue depth, request size and read/write mix. Report tail latency as well as aggregate throughput; a capacity or concurrency specification is not a benchmark.