Input-Output Systems in Operating Systems

Introduction to Input-Output Systems

• Operating System Role in I/O

  • Shares the same device across different processes/users.

  • Abstracts hardware complexities from users: Users do not see how hardware operates.

  • Provides a device-independent interface for uniformity across different devices.

I/O Peripherals Architecture

• Key Components:

  • CPU

  • Memory: Main Memory (including L1 and L2 caches)

  • I/O Buses: Memory Bus (e.g., DDRx), I/O Bus (e.g., PCI)

  • Addresses Range:

  • 0x00…0

  • 0x0ff..f

  • 0x100…0

  • 0x1ff….f

  • 0x200…0

  • 0x2ff….f

Memory-Mapped I/O

• Allows reading and writing to device registers as if they are normal memory.

• User programs typically do NOT have direct access to these operations.

• OS Control: The operating system manages devices, facilitating secure access and control.

• Address Translation:

  • These addresses may not require address translation because:

  • The OS accesses them directly, avoiding enforcement of protection.

  • There is no need for swapping/paging for these addresses.

Disk Device Operations

• Process Overview:

  1. Initial Command:

     • Store command in Command Register with READ_COMMAND.

     • Choose track number via Track Register.

     • Specify sector number in Sector Register.

  2. Load Status:
     Load R, [Status_Reg] cmp R, 0 jeq

  3. Data Transfer:

     • Move data to the target memory.

• Recommendation: Instead of blocking the entire CPU during transfers, employ interrupts to manage tasks efficiently.

DMA (Direct Memory Access) Engine

• Purpose:

  • Offloads the CPU's workload by managing the data transfer between memory and peripheral devices directly.

• Process Overview:

  1. Command configuration for transfer is set (registers same as above plus Memory Address Register).

  2. Disk Request Handling:

     • Use of interrupts to signal completion of transfers.

RAM on Controller

• Serves as a buffer to address the speed mismatch between data streams from the disk and transfers to main memory.

Addressing in DMA

• Address consideration:

  • Physical vs virtual addresses:

  • DMA needs physical addresses as it bypasses MMU (Memory Management Unit) for address translation.

  • The OS programs the DMA ensuring safe handling of addresses.

  • Addresses must be "pinned" in memory, preventing them from being evicted.

I/O Devices Classification

• Block Devices:

  • Store information in fixed-sized blocks, individually addressable.

  • Common Examples: Disks, tapes.

• Character Devices:

  • Deliver or accept streams of characters and are not individually addressable.

  • Common Examples: Terminals, printers, mice, network interfaces.

Principles of I/O Software

• Key Principles:

  • Device Independence: Uniformity ensures that programs operate on any device.

  • Uniform Naming: Names should not directly depend on the type of device.

  • Error Handling: Should occur at the lowest level possible, escalate only if unavoidable.

  • Synchronization:

  • Synchronous (blocking) versus asynchronous (interrupt-driven) operations.

  • Asynchronous is more prevalent, yet synchronous is easier for programming.

I/O Software Architecture

• Layered Approach:

  • Lowest layer (device-dependent): Device drivers.

  • Middle layer: Device-independent OS software.

  • Highest layer: User-level software/libraries.

  • Note: Device drivers and middleware are part of the kernel.

Device Drivers

• Functionality: Handles abstract requests from device-independent software.

• Interface: Consistent interface for all device drivers, including functions like Open(), Close(), Read(), Write(), Interrupt(), Ioctl().

Disk Driver Example

• Semaphore Operations:

  • Open() and Close() methods defined.

  • Read() method programming the device while utilizing semaphores: P(request) for synchronization.

  • Write() and Interrupt handling specifically for disk read events.

Request Handling in Disk System

• Multiple simultaneous requests can complicate order of service.

• Efficiency of serving should consider not just order but seek time optimization.

Disks: Structure and Function

• Physical Structure:

  • Comprises platters with surfaces, circular tracks, and sectors.

  • The assembly allows heads across surfaces to need synchronized positioning on cylinders.

Disk Access Costs

• Actions Involved:

  1. Seek to specific cylinder.

  2. Wait for specified sectors to come under the head.

  3. Transfer data.

• Dominance of Seek:

  • Seek action inherently has the highest cost in terms of time (milliseconds).

Disk Scheduling Algorithms

• First Come First Serve (FCFS):

  • Simple but often inefficient service order (example seek costs calculated).

• Shortest Seek Time First (SSTF):

  • Services closest requests first, minimizing seek time but may lead to starvation.

SCAN Algorithm

• technique where disk head moves in one direction servicing until the end, then reverses.

• Provides bounded traverse distance for the head.

Circular-SCAN (C-SCAN) Algorithm

• Variation of SCAN to ensure uniform wait time.

• Only serves requests in an upward direction and returns to start without servicing in transit.

LOOK Algorithm

• Similar to SCAN but moves only to the last request before reversing; avoids extremes.

C-LOOK Algorithm

• A variant of C-SCAN that also only goes as far as the last request in the direction before reversing.

Device-Independent OS Layer

• Manages device naming and permissions, providing a device independence and error reporting mechanism.

Final Integration in UNIX

• System calls (e.g., open) interact directly with the file system to return device handles.

• Behavior of User Requests:

  • Reference logical block ID for interactions with drivers across all devices.

Conclusion

• The architecture of I/O systems is integral for ensuring efficient communication between the operating system and various hardware components, maintaining efficiency, uniformity, and reliability across operations related to input and output.