I/O Management in Representative Operating Systems

1
I/O Managementin Representative Operating
Systems

2
UNIX SVR4 I/O Management

• Principles
• Each I/O device is associated with a special file
• Special files are managed by the file system
• To read or write to a device, a read or write
  request is made for a special file associated
  with the device
• Types of I/O
• Buffered
• I/O passes through system buffers
• Types of buffers
• System buffer caches
• Character queues
• Unbuffered
• Uses the DMA facility transfer takes place
  directly between the I/O module and the process
  I/O area

3
UNIX SVR4 I/O Management (cont.)
4
UNIX SVR4 I/O Management (cont.)

• Buffer cache
• All I/O operations with the disk (block oriented)
  are handled through the buffer cache
• Data transfer between the buffer cache and the
  user process space uses DMA memory-to-memory
  copy
• Buffer cache management uses three lists
• Free List List of all buffers available for
  allocation
• Device List List of all buffers currently
  associated with each disk
• Driver I/O Queue List of buffers in the process
  of an I/O or waiting for an I/O on a particular
  device
• Implementation
• All buffers are either on the free list or on the
  driver I/O queue list
• No physically separate lists a list is
  maintained as pointers associated with each
  buffer
• Each buffer is associated with a device and
  remains associated with that device even on the
  free list, until reused and associated with
  another device

5
UNIX SVR4 I/O Management (cont.)

• Buffer cache (cont.)
• Operation
• Uses the readers/writers model may be read
  multiple times
• On a reference to physical block X on device Y,
  the OS first checks if the block is in the buffer
  cache
• It searches the Device List for block X on device
  Y
• If entry exists, it contains a pointer to the
  first buffer in the chain
• Block replacement uses a least-recently-used
  algorithm the free list maintains the
  least-recently-used order
• Character queue
• Used for character-oriented devices (e.g.,
  terminals, printers)
• A character queue is either written by the
  process and read by the I/O device or written by
  device and read by the process
• Producer-consumer model is used a character, one
  read, is destroyed

6
UNIX SVR4 I/O Management (cont.)

• Unbuffered I/O
• Fastest I/O, uses DMA between the process space
  and the device space
• Operation
• Process performing unbuffered I/O is locked in
  main memory and cannot be swapped out
• I/O device performing unbuffered I/O is assigned
  to the process for the duration of the transfer
• Types of devices and the buffering used for each
• Disk drives unbuffered I/O or buffer cache
• Tape drives unbuffered I/O or buffer cache
• Terminals character queue
• Communication lines character queue
• Printers unbuffered I/O or character queue

7
Windows 2000 I/O Management

• I/O manager
• Responsible for all I/O for the OS
• Provides uniform interface that all types of
  drivers can call

8
Windows 2000 I/O Management (cont.)

• Cache manager
• Provides caching service in main memory for the
  entire I/O subsystem, serving file systems and
  network components
• It manages the size of the cache for each
  activity
• Provides two services to improve system
  performance
• Lazy write
• Writes done in the cache only and not on disk
• When processor utilization is low, cache manager
  writes changes to disk
• Lazy commit
• Similar to lazy write for transaction processing
• Committed information is written in cache only
• Background process writes information to the file
  system log at a later time

9
Windows 2000 I/O Management (cont.)

• File system drivers
• Treated like any other device driver
• Messages for certain volumes routed to the
  appropriate software driver for that device
  adapter
• Network drivers
• Integrated networking capabilities
• Support for distributed applications
• Hardware device drivers
• Access the hardware registers of peripheral
  devices through entry points in the Executive
  dynamic link libraries
• A set of such routines exists for every platform
  supported by W2K routine names are the same for
  all platforms
• Device drivers portable across different platforms

10
Windows 2000 I/O Management (cont.)

• Asynchronous and synchronous I/O
• Asynchronous mode
• Optimum application performance
• Application initiates I/O operation and the
  calling thread continues execution while the I/O
  operation is queued by the I/O manager and then
  performed
• Application needs to be notified when operation
  is complete
• Synchronous mode
• Application is blocked until the I/O operation
  completes
• Methods for signaling I/O completion for the
  asynchronous mode
• Signaling a device kernel object
• Signaling an event kernel object
• Alertable I/O
• I/O completion ports

11
Windows 2000 I/O Management (cont.)

• Methods for signaling I/O completion for the
  asynchronous mode
• Signaling a device kernel object
• Indicator associated with a device object is set
  when an operation on that object is complete
• The thread that invoked the I/O operation
  continues execution and then checks the indicator
  when it needs the outcome of the I/O operation
• Method is not appropriate for handling multiple
  requests on the same device (e.g., multiple
  actions on a single file)
• Signaling an event kernel object
• Thread creates event for each request
• Thread can then wait on a single or multiple
  requests
• Method is appropriate for handling multiple
  simultaneous I/O requests on a single device or
  file

12
Windows 2000 I/O Management (cont.)

• Methods for signaling I/O completion for the
  asynchronous mode (cont.)
• Alertable I/O
• Uses an Asynchronous Procedure Call (APC) queue
  associated with the thread
• The thread I/O requests and the I/O manager
  places the results in the APC queue for that
  thread
• I/O completion ports
• Used on the W2K server to optimize the use of
  threads
• A pool of threads is available for use no need
  to create a new thread for each new request

13
Windows 2000 I/O Management (cont.)

• Redundant Array of Independent Disks (RAID)
  support
• Hardware RAID
• Separate physical disks combined into one or more
  logical disks by the disk controller
• Controller interface responsible for the creation
  and regeneration of redundant information
• Software RAID
• Noncontiguous disk space combined into one or
  more logical partitions by the fault-tolerant
  software disk driver
• Implemented by the OS and available on the server
  with any set of multiple disks