Structured I/O software; SCAN vs SSTF

Structured I/O software

I/O software is organised in four layers, so that each layer does one job and hides details from the layer above (Tanenbaum).

 User process (read, write, printf)
          |
 +-----------------------------+
 | User-level I/O software     |  libraries, spooling
 +-----------------------------+
 | Device-independent OS soft. |  naming, buffering,
 |                             |  protection, uniform I/F
 +-----------------------------+
 | Device drivers              |  device specific code
 +-----------------------------+
 | Interrupt handlers          |  wake blocked driver
 +-----------------------------+
          |
       Hardware

1. Interrupt handlers: save registers, service the interrupt, unblock the waiting driver, then return. They are kept short and hidden deep in the OS.
2. Device drivers: device-specific code that issues commands to the controller registers, and blocks until the interrupt arrives. One driver per device type.
3. Device-independent software: uniform interface to drivers, naming, protection, buffering, error reporting, allocating dedicated devices, device-independent block size.
4. User-space I/O software: library procedures (e.g. printf) and spooling systems (print spooler).

SCAN vs SSTF (throughput)

For the queue 98, 183, 37, 122, 14, 124, 65, 67 with head at 53, SSTF gives 236 and SCAN (moving towards 0) also gives 236, but SCAN is preferred when load is heavy, as SSTF can starve requests and has unpredictable delay.

Conclusion: SSTF gives slightly better raw throughput for light load, whereas SCAN gives nearly the same throughput with fairness and predictable response time.