Input-output

Transfers of data between a computer and its devices or external environment. An operating-system process can need to wait for an input-output operation to complete, leaving the CPU available for other runnable work. Polled input-output, interrupt-driven input-output and DMA describe different aspects of arranging and observing such transfers.

// Destination: computer-science.bigb

The elapsed delay between a request or event and the corresponding response. A latency model must specify whether setup, waiting, transfer and processing overhead are included.

At constant transfer rate $v$ bytes per second, moving $B$ bytes takes $B/v$ seconds. The size and rate must use compatible units. This term alone does not include seek time, rotational latency or controller setup costs.

The expected value of a latency random variable. If a cache hit takes total time $t_h$, a cache miss takes total time $t_m$, and the cache hit rate is $h$, conditioning gives $\mathbb{E}[T]=h{t}_h+(1-h)t_m$. An overhead already included in either total should not be charged again.

A data storage device accessed in numbered fixed-size blocks. This interface amortizes command overhead and accommodates the physical transfer granularity of the storage medium.

A data storage device that records information magnetically on rotating platters. Accessing a sector can require moving a head to its track, waiting for the sector to rotate beneath it, and transferring the data. These contributions are seek time, rotational latency and data transfer time.^[ref]

The wait for the requested sector to reach the disk head after positioning on its track. For a uniformly distributed arrival phase and rotation period $T_{\mathrm{rot}}$, its expected value is half a rotation. At $r$ revolutions per minute, this is $30/r$ seconds.

The time required to move and settle the disk head over the desired track. It is distinct from seek distance and from waiting for a sector's rotation.

Choose the order of pending input-output requests. Physical head movement, rotational latency, request arrival times and fairness can favour different orders. A seek distance calculation is a simplified model; real seek time is not necessarily linear in track displacement.

Total absolute track displacement along a head route $t_0,t_1,\dots$. Include physical turnaround points in the route if the scheduling policy visits them, even if they do not service a request.

Sweep in one direction through pending requests, reversing at the last pending track rather than travelling to the disk's physical endpoint. It shares the directional sweep of SCAN disk scheduling but can avoid movement into an empty end region.

Sweep in one direction, servicing requests encountered in track order, continue to the physical endpoint, and then reverse. A calculation must state the initial direction and whether movement after the final requested service is included. LOOK disk scheduling instead turns around at the last pending request in that direction.

Serve the pending request whose track is nearest to the current head position, with an explicit tie rule when necessary. This minimizes the next seek distance, rather than the total distance of every future request; a continuing stream of nearby requests can starve a distant request.

Serve requests in arrival order. This preserves queue order but can cause repeated long seeks when successive requests lie far apart.

A device-independent interface expresses requests without exposing each peripheral's control registers or protocol to the client. A device driver translates these requests into device-specific operations. This lets an application retain its interface when the supported hardware changes.

A device register exposes a device's status, control or data to software. Its access semantics can include side effects: a status flag may need an explicit acknowledgement operation before another event is detectable. A device register is not necessarily a processor register. Memory-mapped input-output input-output is one way to access it.

Memory-mapped input-output exposes device registers through addresses accessed by load and store operations. A target-specific C programming language port maps the documented registers to suitably qualified objects and implements any required access barriers. Register layout, access width and acknowledgment semantics belong to the device specification; a register's name alone does not establish its numerical address.