InputOutput Systems

1
Input-Output Systems

• Input-Output (I/O) is perhaps the hardest
  practical task of the OS writer
• I/O devices are very different (mechanically,
  electronically and in purpose and function)
• I/O devices have evolved in many different
  directions and are still evolving rapidly

2
Objectives of I/O Design

• Efficiency and Speed
• I/O devices are very much slower than
  processor/memory and can be a bottleneck
• Very wide range of speeds
• Generality and Device Independence
• We wish to minimise the differences between
  devices to allow a single program to output to a
  printer, a disk or a tape drive, for example.
• Minimise the device-dependent driver code

3
Structure of I/O System
Application Program
User Level
Libraries/DLLs
O/S Input/Output Control System
Operating System Level
Shared I/O Routines - eg Buffering
Device Driver
Device Controller (Hardware/Firmware)
Device Level
Device (Hardware)
4
Devices

• Block Devices
• Transfer information in lumps or blocks
• eg disks and tapes
• May allow random-access
• Character Devices
• Transfer information one character at a time
• Inherently serial
• eg terminal lines, printers
• Some physical devices can be both

5
Buffering

• Most O/S use buffering to improve throughput
  with I/O devices
• Data is exchanged between the device and an O/S
  buffer, and between the buffer and the
  application
• Allows overlap between processing and I/O
• Double buffering allows further gains
• One buffer fills while another is emptied

6
UNIX I/O System features

• UNIX devices appear (by convention) in the /dev
  directory
• I/O devices appear to be files and can be opened,
  closed, read, written, etc.
• Since files in UNIX are streams of bytes, this
  works, although some operations make little sense
  on certain devices
• Device specific features implemented via I/O
  control (IOCTL) system calls

7
UNIX I/O Implementation

• UNIX I/O devices are special files
• Two types - b (block) and c (character)
• Major and minor device numbers correspond to
  device driver and device number (or device type)
  respectively
• Five system calls handle most I/O
• open, close, read, write, lseek

8
Pseudo/Virtual Devices

• UNIX uses various sorts of pseudo/virtual devices
• Some correspond to resources such as memory (eg
  /dev/mem, /dev/kmem)
• Others are convenience features (/dev/null)
• Pseudo-teletypes (ptys) are network-connected
  terminals
• Some devices have multiple virtual devices

9
Device Drivers in DOS

• DOS provided minimal support for devices, which
  relies on the BIOS (Basic I/O System) held in
  read-only memory
• Standard devices include
• con (keyboard, screen)
• com1, com2 (serial ports)
• lpt1, lpt2 (parallel ports)
• A, B (floppy disks)
• C, D, E etc (hard disks or partitions)

10
Windows Device Drivers

• Windows device drivers are implemented as DLL
  files. Advantages include
• shareable code
• optional drivers can be written by third parties
  and installed without modifying Windows
• Windows 95 onwards supports Plug-and-Play (PnP)
  an extended device driver which allows the BIOS
  to initialise/configure a device and for settings
  to be automatically recognised

11
File Systems

• Storage facilities on magnetic disks etc are
  normally provided by a file system
• Many different approaches have been taken some
  O/S (eg VMS) support multiple file-types (eg
  indexed sequential) and are knowledgable about
  the contents
• DOS and UNIX assume files are sequences of bytes
  with no structure

12
Filing System

• Most modern O/S support a hierarchical
  tree-structured filing system
• Directories can contain sub-directories and/or
  files. Top-most directory is root
• Path names identify a file in the hierarchy
• eg /var/adm/syslog/dated/19-Feb-96/mail.log
• Files also accessable relative to current working
  directory using . and ..

13
Directories

• Directories contain information such as
• the file name (up to 255 characters in UNIX)
• the owner (and perhaps group owner) of it
• file access permissions
• size
• creation time, modification time, access time
• pointer to where the information is held on disk
• parent directory in hierarchical systems
• Information known as a link in UNIX

14
Volumes

• A volume is a physical disk (or partition)
• In DOS, volumes are drives C, D, E etc
• In UNIX, volumes are known as file systems
• Partitions are made at the device-driver level
• A volume is prepared by initialising a volume
  data structure to permit creation of files and
  directories on it
• In DOS, this is done with format command
• In UNIX, this is done with mkfs command

15
Allocation of File Space

• File space may be added
• contiguously. This causes fragmentation and
  problems in extending files - they can only be
  extended if free space is available after the end
  of the file. Rarely used in modern systems.
• dynamically in allocation units. Extra file
  space is available on demand and is handled by
  the operating system

16
Allocation of Space in DOS

• DOS allocates file space in clusters.
• NB A maximum of 65,536 clusters/volume
• Cluster size from 512 bytes to 32K
• Used disk clusters are registered in the FAT -
  file allocation table
• FAT is a chain of pointers (null if free)
• Directories point to the appropriate place in the
  FAT table to indicate the start of a file

17
FAT structure
Disk type ID
0
always FFFF
1
0003
2
dir pointer to file 1
0004
3
0006
4
0008
5
dir pointer to file 2
0007
6
FFFF
7
end of file 1
000A
8
0000
9
free
FFFF
A
end of file 2
0000
B
free
0000
C
free
0000
D
free
0000
E
free
0000
F
free
0000
10
free
18
Volume structure for DOS
Boot sector (Bootstrap loader)
Fdisk partition table
FAT table
Duplicate FATs
Root directory
File space
19
Windows NT File System NTFS

• NTFS provides a number of new features
• advanced security
• long (unicode) file names (including spaces)
• improved fault-tolerance through
• recoverable journalled file system
• hot-fixing - bad sectors are transparently
  remapped
• disk striping and/or RAID support
• huge files (up to 264 bytes - 18 x 1012 Mbytes)
• POSIX compatible

20
UNIX File Systems

• Directory information is held in inodes
• File blocks are pointed to by inode pointers
• Inode pointers point to 10 blocks directly plus
• 128 indirect
• 128128 double indirect
• 128128128 triple indirect

21
Directory, Inodes and File
Directory
Details about file
Directory
Pointers
File block 1
File block 2
File block 3
22
UNIX Inode Pointer System
INODE pointer
Direct 0
Data block
Direct 1
Direct 2
Direct 8
Direct 9
ptr to 128
Single Indirect
ptr to 128
Double Indirect
ptr to 128
ptr to 128
Triple Indirect
23
UNIX volume structure
/

• Separate file systems are incorporated into a
  single hierarchical directory with the mount
  command
• A file system can be mounted on any directory
• mount /dev/fd /usr/mnt
• umount unmounts

/usr
/usr/mnt
original root of floppy disk file system
24
Improving a filing system

• Performance
• Blocking and Interleaving
• Cacheing and RAM disks
• Device independence
• Logical file systems which can span disks
• Fault tolerance
• Recoverable file systems
• Disk striping and RAID

25
Blocking and Interleaving

• Blocking
• Disk read time is determined by seek
  timelatencyread/transfer time
• Typical hardware sector size 512 bytes
• Larger block sizes (eg 4K) may improve throughput
  
• Interleaving
• Alters the physical position of data on sectors
  on the disk to improve access to consecutive data

26
Interleaving
1
8
1
8
2
7
5
4
3
2
6
7
4
6
5
3
No interleaving, sectors are adjacent - no time
to transfer sector 1 and set up controller to
read sector 2, so takes 9 revolutions to read 8
sectors.
Interleaved 12 - gives 1 sector gap to
transfer sector 1 and set up controller to read
sector 2. Takes 2 revolutions to read 8 sectors
- more than 4 times faster.
27
Caching

• Disk caching uses a cache (set of buffers) to
  hold recently-accessed disk blocks
• Repeat accesses can be fetched from the cache and
  not the real disk
• May be write back (disk writes delayed) or
  write through (disk updated on writes)
• Write-back risks data loss on power failure

28
Logical File Systems

• Logical files systems, such as Digitals Advanced
  File System, allow a single file system to span
  more than one physical disk
• Add or remove disks at will
• Extend file system sizes as they fill up by
  adding new drives
• Load balancing - moves files around as disks are
  added so that all drives get equal use
• Allow several file systems to share 1 drive

29
Fault Tolerance

• Fault Tolerance - literally the ability to
  tolerate (and recover) from hardware or software
  errors
• Software techniques include
• transaction based file systems
• hot fixing - data on bad sectors is automatically
  moved to a new location and old one marked
• Hardware techiques include disk striping, RAID

30
RAID

• Redundant Array of Inexpensive Disks
• expensive RAIDs are R A of Independent Ds!
• RAID improves
• performance and/or
• reliability (fault tolerance
• RAID systems combine a number of disks into a
  single file system

31
RAID-0

• RAID-0 is also known as disk striping
• R-0 can have any number of disks
• Blocks are assigned to disks in rotation - first
  block to first disk, 2nd to 2nd, etc
• Improves performance a read of a number of
  consecutive blocks causes several disks to be
  used - speeds transfers
• Doesnt help reliability (makes it worse!)

32
RAID-0
Block 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
Disk 1
Disk 2
Disk 3
Disk 4
33
RAID-1 Mirroring

• Each drive is mirrored so each block is recorded
  on at least two disks
• Fault tolerant One disk can fail
• Improved performance choice of drives to read
  from, so throughput similar to R-0
• eg block(b)1/disk(d)1, b2/d3, b3/d2, b4/d4
• Main disadvantage is that 1/2 the capacity of the
  storage array is used in mirroring

34
RAID-1
disk from which that block is read
round-robin all writes go to both mirrored
drives
Block 1
1
2
2
3
3
4
4
5
5
6
6
7
7
8
8
9
9
10
10
11
11
12
12
13
13
14
14
Disk 1
Disk 2 (disk 1 mirror)
Disk 3
Disk 4 (disk 3 mirror)
35
RAID-3 Striping with parity

• Note RAID-2 exists but is never used!
• RAID-3 is similar to RAID-0, but reserves one
  disk for parity
• Parity is calculated as the exclusive-OR of the
  data blocks
• XOR parity is reversible - given the failure of
  any disk one can calculate the information that
  is missing

36
RAID-3
Block 1
2
3
1 xor 2 xor 3
4
5
6
4 xor 5 xor 6
7
8
9
7 xor 8 xor 9
10
11
12
10 xor 11 xor 12
13
14
15
13 xor 14 xor 15
16
17
18
16 xor 17 xor 18
19
20
21
19 xor 20 xor 21
Disk 4 (Parity disk)
Disk 1
Disk 2
Disk 3
37
RAID-3

• RAID-3 advantages/disadvantages
• It normally delivers the same read performance as
  RAID-0 (unless we need to use parity disk)
• Any one disk can fail without affecting data
  which can be calculated from the parity (although
  reads are much slower)
• The parity disk is overworked, being used on
  every write operation
• Storage loss is modest - commercial systems use a
  single parity disk for up to 7 data disks

38
RAID-5

• NB Raid-3 and Raid-4 are almost identical
• RAID-5 is similar to RAID-3, but with the parity
  block rotated around the disks
• All disks get equal use whether reading or
  writing
• Practical RAID-5 systems may allow hot swapping
  or stand-by spare arrangements
• Read about RAID on the web for other issues

39
RAID-5
Block 1
2
3
1 xor 2 xor 3
4 xor 5 xor 6
4
5
6
7
7 xor 8 xor 9
8
9
10
11
10 xor 11 xor 12
12
13
14
15
13 xor 14 xor 15
16 xor 17 xor 18
16
17
18
19
19 xor 20 xor 21
20
21
Disk 1
Disk 2
Disk 3
Disk 4