Topics

• Device management: Characteristics of serial and parallel devices; abstracting device differences; buffering strategies; direct memory access; recovery from failures.
• File systems: Fundamental concepts (data, metadata, operations, organization, buffering, sequential vs. nonsequential files); content and structure of directories; file system techniques (partitioning, mounting and unmounting, virtual file systems); memory-mapped files; special-purpose file systems; naming, searching, and access; backup strategies.

Device Management

• Major device characteristics
  • Data volumes
  • Transfer rates
  • Latency
  • Possibility of error
  • Modes of access: serial vs. parallel
    • Serial access provides simplicity
    • Parallel access gains speed, but needs to deskew
    • Distinctions becoming blurred
  • Character vs. block oriented
    • Serial access natural for asynchronous bit and character-oriented devices
      • Terminals
      • Each character is handled independently
    • Parallel access favors synchronous block orientation
      • Disks
      • Tapes
    • Fast serial access can be used for synchronous block-oriented devices
• Device interfaces and controllers
  • Provide a translation between external physical states and internal logical states
  • Abstract devices
  • Logic has been moving into the devices
  • Manages data and metadata (state)
    • Establishes state for transfer
    • Transfers data
    • Checks state after transfer
  • I/O port control
    • Special instructions (IN, OUT) access device registers
    • Data may be transferred through registers or via DMA
    • May poll or be interrupt driven
  • Memory-mapped I/O (PDP-11)
    • Certain memory addresses mapped to device registers
    • Data may be transferred through registers or via DMA
    • May poll or be interrupt driven
  • Mixed strategies (PCs, video devices)
  • Programmed I/O (PIO)
  • Direct memory access (DMA)
  • Interrupts
    • May be used both with PIO and with DMA
    • Interrupt vector
    • Stack to be used
    • Precise interrupts
    • Imprecise interrupts
• I/O software
  • OS Interrupt management
    • Kernel level code
    • May be purely hardware
    • Directs particular interrupt events to particular drivers
    • Manages context (memory maps, registers, stacks)
    • Manages nesting of interrupts
  • Device drivers
    • May include interrupt management
    • Usually kernel level code
    • May be character device, block device, both or neither
    • Functions: read, write, position, set/get status
    • Must be careful about blocking
    • Abstract classes of devices (device independence)
    • Manage errors
    • Manage asynchronous transfers (backgrounding)
    • Buffer data
    • Share devices among data streams
    • Share data streams among devices
    • Share interrupts
  • OS device request management
    • Queue requests for devices
    • Manage blocked requests
    • Error handling
    • Reblock data
    • Common services for drivers
    • Access control
      • Shared devices
      • Allocatable devices
  • Run time libraries (run time environments)
    • Language-dependent interface to I/O systems
    • May depend on file systems
    • Open, close, position I/O streams
• Typical devices
  • Block-oriented
    • Disks
      • Magnetic disks, WORM disks (CDs), CD-R, CD-RW, DVD
      • Heads, surfaces, tracks, cylinders
      • Moderate latency, fast transfer rates
    • Tapes
      • Digital vs. analog
      • Low cost, but capacities falling behind disk capacities
      • Very high latency, moderately fast transfer rates
      • Contacting medium, reliability issues
    • Some terminals, some printers
  • Character-oriented
    • Keyboards, mice, many terminals, many printers
  • Graphics devices
    • Text vs. vector vs. raster graphics
    • Often treated as bit-mapped memory
    • Messages oriented pseudo-devices created for full GUI
    • X-windows vs. thin clients

File Systems

• Files as abstract containers of information with identifiers (names)
  • Extends the concepts of variables, arrays and data structures
  • May simply be an organizational concept for the RAM of a computer
  • May extend the available storage space beyond the RAM of a computer to peripheral devices
  • May allow information to be shared among multiple programs
    • Sequentially (one program writes data, complete, then another reads it)
    • In parallel -- raises questions of ownership, interlocks and coherence of data
  • A file may have fixed or dynamic organization
  • A file may have the lifetime of the program or job that creates it
  • A file may persist beyond the lifetime of the program or job that creates it
• Most useful when organized into hierarchies (some files are directories of other files)
• Allows data to be truly independent of particular devices (disk, tapes, ...), and sometime of particular computers
• Naming conventions
  • Range from trivial (a number) to complex (internet URL's)
  • May identify a particular container for data independent of the data that is stored within it (e.g. a system swap file)
  • May identify a particular instance of data (e.g. the executable of a program I just compiled)
  • May identify a unit of information independent of where it is stored (e.g. the 7.2 release of RedHat Linux)
  • Identify which file, and may provide attributes and relationships to other files
  • Case sensitivity (Windows -- case-insentitive, Unix -- case sensitive)
  • Valid styles of names (8.3, long names, blanks)
  • Specify attributes of files by portions of their names (Windows extensions, unix hidden files)
• Mapping information into abstract file organizations
  • Sequential abstract file organizations
    • Information organized into sequences of characters or records
    • Information always read from the beginning
    • Information always written at the end
    • May be organized in terms of fixed length records or more complex record types
  • Direct access file organizations
    • Information organized into fixed sized bins
    • Information may be read from/written into any bin
    • Records need not be the same structure as the bins
  • Keyed access file organizations
    • tree structures combining sequential and direct access methods
• Mapping files onto disk blocks
  • Need not follow the file orgaization but needs to support the relevant access methods
  • Multiple record and file types
  • Contiguous vs. non-contiguous files
  • Extensible vs. non-extensible files
  • Sequential vs. direct access files
• Types of files
  • Some files contain data for ordinary programs
  • Some files contain data which organizes other files (directories)
  • Some files provide access to devices (character special, block special)
  • Some systems use external attributes or naming to identify the type of contents of the file
  • Some system parse the file to recognize the contents (magic numbers)
• File Operations
  • Create
  • Remove (delete the directory entry, delete the file contents)
  • Attach a program to a file (open)
  • Disassociate a program from a file (close)
  • Establish position (seek) and attributes relative to the file
  • Transfer data from a file to a program (read, execute)
  • Transfer data from a program to a file (write, append)
  • Modify name and other attributes of a file
• Access controls
  • Mechanisms needed to decide which processes may perform which operations
  • May be controlled by user, by process, by time, by operation, etc.
  • Unix convention is user, group, other, read, write, execute
  • Many systems provide complex access control lists
  • Many system provide interactive record and file locking mechanisms
• Directory structures
  • Hierarchical directories allow many files to have the same name
  • Paths specify which branch(es) of the structure are to be used
  • Remote mounts allow different systems to share portions of directory trees
  • Loops in the tree structure are a disaster