Data Link Layer Services and Protocols

1
Data Link Layer Services and Protocols

• Arzad A. Kherani
• (alam_at_cse.iitd.ac.in)
• Dept. of Computer Sc. And Engg.
• Indian Institute of Technology Delhi

2
Outline

• Frame encoding
• Error detection and recovery
• Data Link Protocols
• Protocol analysis
• Performance
• Verification for correctness

3
Data link services

• Operates between two neighboring devices
• Provides a capability for higher-layer entities
  to send packets
• A packet is a sequence of bits, with
  well-identified start and end
• The packet is itself encapsulated into a frame,
  adding to it header and trailer

4
Data link services (2)
5
Data link services (3)

• There is one data link for each physical link
• In a router, for instance

6
Data link services

• Connection-less service
• Un-acknowledged service
• Useful in case of low errors, and for real-time
  applications
• Acknowledged service
• Used in wire-less networks
• Frames that are not acked may be re-sent
• Connection-oriented service
• acknowledged
• Ensures delivery of frames

7
Framing

• Link  layer  packet    frame
• Problem  how  to  recognize  beginning  and 
  end  of  frame?
• Three  methods
• byte  counting  (DDCMP)
• bit  stuffing  (X.25  Level  2,  802.x)
• byte  stuffing  (BISYNCH,  IMP-IMP)

8
Framing

• Solution based on counting bits/characters

9
Framing

• Solution based on flags (01111110) and bit
  stuffing

Original bit string
Bit string, suitably bit-stuffed
Received, interpreted bit string
10
Bit Stuffing (2)

• Frame  beginning  and  end  marked  by  special 
• bit  string  ( 01111110)
• If    5  1's  in  data  to  be  sent,  sender 
  inserts  0
• If  receiver  sees  5  1's  check  next  bit(s)
• if  0,  remove  it  (stuffed  bit)
• if  10,  end  of  frame  marker  (01111110)
• if  11,  error  (7  1's  cannot  be  in  data)

11
Framing

• Solution based on flag characters, byte stuffing
• Special  characters  used  for  control

12
Byte Stuffing Problems

• Dependence  on  fixed  character  set
• Must  examine  every  byte  of  data  on 
  sending 
• and  receiving  (insert  /  remove  DLE)
• Was  used  widely  in  IBM  bisynch  (at 
  9600bps)

13
Error detection and recovery

• Two approaches
• error correction codes
• quick, but ineffective in several cases
• temporary dislocation
• frame size is large, and error rate is high
• burst errors
• expensive
• error detection and recovery using
  re-transmission
• the preferred solution today
• efficient

14
Error detection

• Hamming distance between pair of codes
• let message of length m ? 2m distinct messages
• with r redundant bits, codeword is of length n
  m r
• hamming distance between codes x, y
• no. of bits in which x and y differ
• Hamming distance for a code
• consider n dimension space, with 2m codewords
• hamming distance for code (or coding scheme)
• minx, y (no. of bits in which x and y differ)

15
Hamming codes

• Hamming distance for code based on parity bit is
  2
• resulting capability detect 1 error, correct 0
  errors
• Result
• to detect d errors, the Hamming distance must be
  d1
• to correct d errors, the Hamming distance must be
  at least 2d 1

16
Hamming codes (2)

• Consider 7 bit data, 4 redundancy bits, codeword
  is 11 bits
• message bits are numbered 3, 5, 6, 7, 9, 10, 11
• redundancy bits are numbered 1, 2, 4, 8
• check bit 1 checks error in bits 1, 3, 5, 7, 9,
  11
• check bit 2 checks error in bits 2, 3, 6, 7, 10,
  11
• check bit 4 checks error in bits 4, 5, 6, 7
• check bit 8 checks error in bits 8, 9, 10, 11
• or (it should be)
• 0 b1 b3 b5 b7 b9 b11
• 0 b2 b3 b6 b7 b10 b11
• 0 b4 b5 b6 b7
• 0 b8 b9 b10 b11

17
Error detection using block codes

• Block of data is re-written as a matrix, say 4 x
  8
• last row is parity bits
• bits are transmitted row-by-row, including parity
  bits
• code is capable of detecting a burst of errors of
  length n

18
Polynomial codes

• Also known as Cyclic Redundancy Codes (CRC)
• Note, all arithmetic is modulo-2

19
Polynomial codes (2)

• Error detection capability depends upon G(x)
• 1 bit error ? R(x) T(x) xi
• R(x)/G(x) 0 iff E(x)/G(x) 0
• E(x)/G(x) ? 0 if G(x) has 2 or more terms
• ? single errors can always be detected
• Similarly, two errors can always be detected if
  G(x) does not divide xk 1
• Again, if G(x) is divisible by x1 then an odd
  number of errors can be detected
• Polynomial codes with r CRC bits will detect all
  bursts of length r or less
• etc., etc.

20
Polynomial codes (3)

• International, IEEE standards
• IEEE 802 standard
• G(x) x32 x26 x23 x22 x16 x12 x11
  x10 x8 x7 x5 x4 x2 x1 1
• it is capable of detecting bursts of length up
  to 32, and all odd number of errors, and even no
  of error with high probability

21
Error recovery

• Error detection, followed by re-transmissions,
  etc.
• Efficient
• Simultaneously address problem of flow control

22
Elementary data link protocols

• Broad objective of data link protocol
• Error-free, loss-free, duplication-free and
  in-sequence transfer of user data packets between
  network entities
• flow-controlled
• transfer user data packets in both directions

23
Data link modules
24
Data link modules

• We consider several protocols. But to begin with
  we consider moving data in one direction, only

25
Data link protocol

• Assumptions
• errors during transmission
• processing capacity at receiver end
• buffer capacity at receiver end
• whether the underlying physical channel is half-
  or full-duplex
• we will make nice assumptions to begin with, but
  move towards realistic assumptions later

26
Simple data link protocol Utopia

• Assume no errors, infinite processing capacity
  and buffer space at receiver end, and half-duplex
  channel

Note, F_i Data link frame, containing data
packet, i
27
Simple data link protocol Utopia (2)

• The senders end

28
Simple data link protocol Utopia (3)

• The receivers end

29
Definitions of data types/structures
30
Definition of procedures
31
Simple data link protocol Utopia (4)
32
Flow control problem and its solution

• Flow control --gt limit the rate at which a sender
  can send data to one which is consistent with the
  receivers ability to process incoming data
• Two approaches to solving it
• rate-based determine the minimum rate at which
  receiver can process incoming data
• feedback based send more data as when receiver
  can handle more data

33
Stop and wait protocol

• ?

• Assume no errors, FINITE processing capacity,
  FINITE buffer space at receiver end, and
  half-duplex channel

Note, F_i Data link frame, containing data
packet, i
34
Stop-n-wait protocol (2)
35
PAR protocol for noisy channels

• PAR protocol addresses
• flow control
• noisy channel
• based on positive acks and re-transmissions

36
PAR protocol
37
PAR protocol

• Data Frames are suitably numbered 0, 1, 0, 1,
• Acks are not numbered

38
PAR protocol (sender)
39
PAR protocol (receiver)
40
PAR protocol
41
PAR protocol

• If the underlying physical layer is full-duplex,
  then the protocol fails

42
Alternating bit protocol

• PAR protocol, with numbered acks

F_0
Pkt 1
Pkt 1
Ack_0
Pkt 2
F_1
F_1
Pkt 2
Ack_1
Pkt 3
F_0
Pkt 3
Ack_0
F_0
Ack_0
Sender, A
Receiver, B
43
Alternating bit protocol
44
Alternating bit protocol
45
Alternating bit protocol

• Two possibilities

46
Alternating bit protocol

• Use link A ? B to carry data from A to B, and
  acks from B to A and use link B ? A to carry data
  from B to A, and acks from A to B
• Piggyback acks onto data frames, if one has data
  to send
• Else, just an ack
• Redundant acks are OK
• An ack takes the form I am waiting to receive
  data frame no. X
• Introduce a field for frame type, data frame or
  ack frame

47
Performance of PAR protocol

• Link utilization
• Utilization L / (LbR), where
• L size of data frame
• b data rate
• R is round-trip delay
• For a satellite channel, let L 10K bits, b
  100 Kbps, R500ms
• Utilization is 10K / (10K 50K) 1/6
• For a fibre-optic channel, let L 10K bits, b
  100 Mbps, R 1ms
• Utilization is 10K / (10K 100K) 1/11
• DelayBW product is the key
• Let sender send many data frames before it
  receives an ack

48
Pipelining

• Problems arise when one or more packets are lost

49
Pipelining, with error recovery

• Two approaches to recover from loss of data
  frame
• go-back n
• selective repeat

50
Pipelining, with error recovery

• Go-back n scheme

51
Pipelining, with error recovery

• Selective repeat

52
Sliding window protocols

• Each data frame is sequentially numbered 0
  through 2n-1
• Sender maintains transmit window indicating
• which data frames can be sent
• which data frames have been sent
• receiver maintains a Receive window
• which data frames can it receive
• which data frames have been received

53
Sliding window protocols

• Transmit window
• size 4, waiting for acks for data frames 1, 2,
  3, frame 4 not sent
• Receive window
• size 3, ack for frame 1 sent, data frame 3
  received, waiting for frames 2 and 4

54
Sliding window protocols

• Assuming sequence numbering 0 .. 7
• Go back n protocol
• Transmit window is size, say 7
• receive window is size 1
• selective repeat protocol
• transmit window is size 4
• receive window is size 4

55
Go back n protocol declarations
56
Go back n protocol more declrations
57
Go back n protocol initializations
58
Go back n protocol (loop)
59
Go back n protocol event processing
60
Go back n protocol event processing
61
Go back n protocol event processing
62
Go back n protocol

• Buffer requirements
• at senders end N-1, where seq no is 0 .. N-1
• at receivers end 1
• Go-back n works well when error are infrequent
• Because several frames need to be re-transmitted
  when an error occurs

63
Select Repeat Protocol

• Sequence numbering 0 through n-1
• Transmit window size is n/2
• Receive window size is n/2
• Receiver buffers each correctly recd data frame,
  but does not deliver it to the layer 3
• Receiver sends a NACK frame when it suspects loss
  of a data frame
• But makes sure that a NACK is not repeated

64
Select Repeat Protocol
65
Select Repeat Protocol
66
Select Repeat Protocol
67
Select Repeat Protocol
68
Protocol Analysis

• Performance
• Verification

69
Protocol Performance

• Performance
• Delay
• Efficient use of available bandwidth

70
Channel utilization

• It can be shown for stop-and-wait protocol
• U F1 F2 F3
• where
• F1 D/ (HD)
• F2 (1-E)(HD) (1-E)D
• F3 (HD) / (HDCT)
• Above,
• D, H are resp. length of data and header (or
  Ack)
• E is bit-error-rate (BER)
• C is channel capacity
• T is timer interval

71
Channel utilization

• Channel utilization in stop-and-wait protocol
  is maximum when
• Doptimum ?(HCT)/E

72
Channel utilization

• Channel utilization in sliding-window protocols
  is complex. Consider
• No error, large Tx window
• No error, small Tx window
• With possibility of errors, large Tx window
• With possibility of errors, small Tx window

73
Channel utilization

• What is a large enough window size?
• W gt 1 2 C I / (DH)
• where I propagation delay interrupt
  handling time

74
Channel utilization

• Channel utilization in sliding-window when
  there is no error, large Tx window
• U D / (DH)
• Channel utilization in sliding-window when
  there is no error, but small Tx window
• U D/(DH) W/(1 2 C I / (DH))

75
Channel utilization

• Channel utilization in sliding-window when
  there may be errors, but Tx window is small
• U D/(DH) (1-L)
• where L is the probability that a frame or an
  ack to it is lost
• Channel utilization in sliding-window when
  there may be errors, but Tx window is small
• U D/(DH) (1-L) W/(1 2 C I / (DH))

76
Channel utilization

• CI is also the delay-BW product,
• And CI/F is the delay-BW product in terms of no
  of frames

77
Protocol Verification

• Verification
• Formal specification
• Verification of protocol design
• Conformance of an implementation to a given
  design

78
Protocol Verification

• Two parts
• protocol modeling
• verification
• Look upon the communicating entities as ONE
  finite state machine
• Use two different, although equivalent, ways to
  model the machines
• State transition diagrams
• Petrinet based model

79
Protocol verification model using state
transitions

• Consider an example stop-n-wait protocol
• Consider the sender-receiver pair, together with
  channels as ONE single system
• Focus on significant states, not intermediate
  states encountered while executing commands or
  program instructions

80
Protocol Verification model using state
transitions

• Stop-n-wait protocol components, and their states
• Sender sending 0 or 1
• Receiver receiving 0 or 1
• Half-duplex channel carrying frame 0 or 1, or
  Ack, or -
• Acks are not numbered
• Total no of states is 16, of which 6 are
  unreachable
• These are states that correspond to sender and
  receiver waiting for next event to occur
• Further,
• Consider initial state
• Any enabled transition may take place next
• Transitions take place at any time

81
Protocol Verification model using state
transitions
82
Protocol Verification

• Can a protocol drop a user packet?
• Protocol never delivers packets in two data
  frames, numbered 0, without delivering packet
  contained in a data frame numbered 1
• Or, machine does not make two transitions
  numbered 1, without an intervening transition
  3
• There does not a path with two edges
  corresponding to transition 1 without an
  intervening edge transition 3

83
Protocol Verification

• Can a protocol drop a sequence of two user
  packets?
• Or, sender should not state 1 twice while
  receiver does not change its state in between

84
Protocol modeling using Petrinets

• A Petrinet, with
• places set of possible states
• tokens define the current state
• transitions, input and output arcs
• firing of transitions
• conditions, and resulting re-distribution of
  tokens

85
Producer/consumer model

• Model of a producer- consumer system with one
  buffer, using Petrinet

86
Stop-n-wait protocol model
87
Stop-n-wait protocol model

• Petrinet is formally described by its
  transitions
• 1 BD --gt AC
• 2 A --gt A
• 3 AD --gt BE
• 4 B --gt B
• 5 C --gt
• 6 D --gt
• 7 E --gt
• 8 CF --gt DF
• 9 EG --gt DG
• 10 CG --gt DF
• 11 EF --gt DG

88
Stop-n-wait protocol model

• Using the formal specs one can formally argue
  about properties that are (are not) satisfied
• e.g. consider all possible sequences of
  transitions. Then are there two 10 transitions
  without an intervening transition 11?
• How would ensure that two consecutive packets are
  not lost

89
Example Data Link Protocols

• HDLC (bit-oriented, high-level data link control)
• Connection establishment, release, data transfer,
  and even connection reset

90
Example Data Link Protocols

• PPP (point-to-point protocol)
• Mainly used router-to-router and dial-up
  connections
• Connection establishment and release a data link,
  data transfer, and even connection reset
• Help establish network layer (IP) addresses
• Over LANs, the protocol is typically Ethernet

91
Thanks