Operating Systems (AD-405) Unit 5 Notes - RGPV

How unit 5 is examined

This unit covers how the OS drives devices (polling, interrupts, DMA, buffering, drivers) and introduces network, distributed and multiprocessor systems; the distributed-systems topic carries almost all the marks.

Principles and Programming

Not asked since 2022

Definition. I/O principles describe how the OS hides device hardware behind a uniform interface, using a device controller for each device and a driver for each controller.

Key points.

A device controller is the electronic part that operates the device and has status, command and data registers.
The CPU programs I/O by writing commands into controller registers, either through I/O ports or memory-mapped I/O.
Devices are block devices (disk, fixed-size blocks) or character devices (keyboard, stream of bytes).
The OS should give device independence and uniform naming, so programs never depend on the hardware.

Input/Output Problems

Definition. I/O problems are the difficulties caused by devices being far slower and more varied than the CPU.

Key points.

Speed mismatch: a CPU executes millions of instructions while a disk or printer completes one operation, so the CPU would sit idle.
Devices differ in data rate, unit of transfer, data format, and error conditions, which makes one uniform interface hard.
Devices are shared, so requests must be scheduled to avoid conflict and deadlock.
Buffering, interrupts, DMA and spooling are the standard solutions to the speed mismatch.

Different I/O operations: Program Controlled

Definition. In program-controlled (programmed) I/O the CPU itself starts the transfer and repeatedly tests the device status until it is ready; this is called polling or busy waiting.

Key points.

The CPU issues the command, then loops reading the status register until the ready bit is set.
It is simple and needs no extra hardware.
It wastes CPU time because the CPU does no useful work while waiting.
Polling is acceptable only when the device is fast or the wait is short.

Interrupt Driven

Definition. In interrupt-driven I/O the CPU starts the transfer and continues other work; the device raises an interrupt when it is done and an interrupt handler moves the data.

Point	Interrupt driven	Polling
CPU use	Free between events	Busy waiting
Overhead	Context switch per event	Loop per check
Latency	Fast response	Depends on poll interval
Best for	Rare events	Very frequent events

Key points.

A single serial port sends few characters, so an interrupt per character is cheap and the CPU stays free between them.
A front-end processor (terminal concentrator) has data almost always ready, so interrupts would fire constantly; polling costs less than thousands of context switches.

Asked: [7 marks] (Jun 2025) Why might a system use interrupt driven I/O to manage a single serial port but polling I/O to manage a front end processor, such as a terminal concentrator.

Concurrent I/O

Definition. Concurrent I/O lets I/O overlap with computation; Direct Memory Access (DMA) is the hardware that does it by moving a block between device and memory without the CPU.

Key points.

The CPU gives the DMA controller the device, memory address, and byte count, then carries on.
The DMA controller transfers the whole block and raises one interrupt at the end.
It steals memory cycles from the CPU (cycle stealing) but far fewer than programmed I/O.
DMA gives one interrupt per block instead of one per byte.

Asynchronous Operations

Definition. Asynchronous (non-blocking) I/O returns control to the caller at once and the process is told later when the transfer completes.

Key points.

In synchronous (blocking) I/O the process waits until the operation finishes.
In asynchronous I/O the process continues and is notified by a signal, callback, or flag.
It improves overlap of computation and I/O but the program is harder to write.
Blocking waits for the result; asynchronous continues and gets notified.

Logical structure of I/O function

Definition. I/O software is organised in layers, each hiding the details of the layer below.

Key points.

From top to bottom: user-level I/O software, device-independent OS software, device drivers, interrupt handlers, hardware.
Interrupt handlers save state, service the device and wake the blocked driver.
Device drivers hold the device-specific code that talks to the controller registers.
Device-independent software does naming, protection, buffering, error reporting and allocation.
Layering gives device independence: changing a device only changes its driver.

I/O Buffering

Definition. Buffering keeps data in a memory area while it moves between a device and a process, to smooth the speed mismatch. Spooling queues output (or input) for a dedicated device, such as a printer, on disk.

Key points.

Single buffer: the device fills one buffer while the process uses the previous data.
Double buffer: the device fills one buffer while the process empties the other, giving more overlap.
Circular buffer: many buffers in a ring for bursty transfers.
Spooling lets many jobs "print" concurrently into disk files; a spooler daemon sends them to the printer one by one, so the device appears shared.
Buffering holds data for one transfer; spooling queues whole jobs for a non-shareable device.

Asked: [7 marks] (Nov 2023) Explain the concept of buffering and spooling.

Kernel I/o Subsystem

Definition. The kernel I/O subsystem provides the services built on drivers and hardware.

Key points.

I/O scheduling orders requests to improve average waiting time and fairness.
Buffering and caching keep data in memory; a cache holds a copy of data for faster access.
Spooling and device reservation manage devices that cannot be shared.
Error handling retries transient failures and reports permanent ones.
The subsystem also provides protection: users reach devices only through system calls.

Introduction to Network, Distributed and Multiprocessor Operating Systems

Definition. A network OS lets users log in to and use remote machines that know they are networked; a distributed operating system makes a collection of independent computers appear to the user as a single system. A multiprocessor OS manages several CPUs sharing one memory.

Key points.

Network OS: users are aware of machines and copy files explicitly; distributed OS: the system decides where work runs.
Design issues: transparency (access, location, migration, replication, concurrency, failure), communication (message passing, RPC), reliability, scalability, heterogeneity and security.
Transparency advantages are ease of use, resource sharing and location independence; disadvantages are overhead, hidden failures, security exposure and loss of user control.
Failure types: crash (node halts), omission (message or response lost), timing (response too late), Byzantine (arbitrary or malicious behaviour), network partition.
DFS benefits over a centralised file system: scalability, replicated files for fault tolerance and availability, parallel access for performance, and sharing from any node.
On a totally reliable network, drop retransmission, timeouts and message-loss handling; keep naming, migration, performance and node-crash tolerance.
RPC: a client calls a remote procedure like a local one; the client stub marshals the arguments into a message, the server stub unmarshals them and runs the procedure, and the result comes back the same way. Example: NFS.
Process migration: freeze the process, copy its state (registers, address space, open files), and resume it on the target node.

Process migration method.

Step 1: Select process and destination; freeze the process.
Step 2: Capture state: registers, address space, open files, signals.
Step 3: Transfer the state to the destination.
Step 4: Same OS: restore directly. Different OS or architecture: convert data formats and system calls.
Step 5: Resume; forward messages, redirect open files.

Answer frame. Open with the definition of a distributed OS; for design issues develop point 2 in order transparency, communication, reliability, scalability, heterogeneity, security; for transparency, give advantages then disadvantages then types; for failures, define the three then the impact on reliability; for migration draw the steps above; close with the trade-off between transparency and performance.

Pitfall: Answering "failures" with only crash; name three with a line each.

Asked: [7 marks] (Nov 2023) Discuss the design issues of distributed operating system. Asked: [7 marks] (Jun 2024) What are the advantages and disadvantages of making the computer network transparent to the user? Asked: [7 marks] (Jun 2024, Jun 2025) List three possible types of failures in a distributed system. Asked: [7 marks] (Jun 2024) What are the benefits of a DFS when compared to a file system in a centralized system? Asked: [7 marks] (Jun 2025) What aspects of a distributed system would you select for a system running on a totally reliable network. Asked: [7 marks] (Jun 2025) Describe a method for process migration across different architecture running: (i) the same operating system, (ii) different operating system. Asked: [3 marks] (Jun 2023) Write short notes: Remote Procedure Call (RPC)

I/O management in UNIX & Windows

Definition. UNIX treats every device as a file, while Windows uses a layered I/O manager with drivers.

Key points.

UNIX devices appear in /dev as block or character special files, using the same open, read, write and close calls.
UNIX uses a buffer cache for block devices and character queues for terminals.
The Windows I/O manager passes I/O request packets (IRPs) down a stack of drivers.
Windows supports asynchronous I/O and plug and play drivers.
UNIX: "everything is a file"; Windows: I/O manager plus driver stack.

Last-minute revision

Polling wastes CPU; interrupts free it; DMA gives one interrupt per block.
Serial port: interrupts; terminal concentrator: polling.
Buffering is temporary memory; spooling queues jobs on disk.
Double buffering overlaps device and process.
Distributed OS gives a single-system image.
Failure types: crash, omission, timing, Byzantine.
RPC uses client and server stubs with marshalling.
Layers: user software, device-independent, driver, interrupt handler, hardware.
UNIX devices are files in /dev.
Reliable network: no retransmission needed.

Memory hooks

CO-TB: Crash, Omission, Timing, Byzantine.
DMA = one interrupt per block, not per byte.
Spooling = Simultaneous Peripheral Operations On-Line.
Transparency: Access, Location, Migration, Replication, Concurrency, Failure.

Coverage checklist

Principles and Programming: definition, controllers, device independence.
Input/Output Problems: speed mismatch, variety.
Different I/O operations: Program Controlled: polling.
Interrupt Driven: Jun 2025 serial port vs concentrator.
Concurrent I/O: DMA.
Asynchronous Operations: blocking vs non-blocking.
Logical structure of I/O function: layers.
I/O Buffering: Nov 2023 buffering and spooling.
Kernel I/o Subsystem: scheduling, caching, spooling.
Introduction to Network, Distributed and Multiprocessor Operating Systems: Nov 2023, Jun 2024 (three), Jun 2025 (three), Jun 2023 RPC.
I/O management in UNIX & Windows: device files, IRPs.

Account

Operating Systems (AD-405) - Unit 5 Short Notes

How unit 5 is examined

Principles and Programming

Input/Output Problems

Different I/O operations: Program Controlled

Interrupt Driven

Concurrent I/O

Asynchronous Operations

Logical structure of I/O function

I/O Buffering

Kernel I/o Subsystem

Introduction to Network, Distributed and Multiprocessor Operating Systems

I/O management in UNIX & Windows

Last-minute revision

Memory hooks

Coverage checklist