Link Layer: Motivation

1
Link Layer Motivation
B
A
Message M

• Problem Given a message M at a node A consisting
  of several packets, how do you send the packets
  to a neighbor node B
• Neighbor A node attached to the same link
• Link can be point-to-point or broadcast
• Link can be guided media (a copper, coax, fiber
  wire) or unguided media (wireless)

2
Link and Physical Layers

• This communication problem is handled by 2
  protocols
• A Link Layer (LL) that sits on top of the
  physical layer (PL) and deals with
• Packet Encapsulation, Mux/Demux
• Framing Detecting frame boundaries
• Error Detection/Recovery Detecting corrupt
  frames
• Media Access Control (if the link is multi-access
  or broadcast)
• Reliable delivery, flow control? Optional
• A Physical Layer (PL)
• deals with encoding/decoding bits of a frame
  to/from the link

3
Network Interface Card (NIC)

• LL in part, PL in total are implemented in NIC
• Ethernet card, 802.11 card,
• NIC is semi-autonomous
• Listens to the link independent of the CPU
• Talks to CPU after reception of a new frame
• CPU talks to the card to send a frame
• Has connectivity to both the I/O bus and the
  network link

4
Link Layer Service Interface

• Every Link Layer must export a service interface
  to the layers on top of it
• Unreliable frame service
• Corrupt frames are not recovered, frame sending
  order may not be preserved
• OK when error rate is low so packet recovery is
  left to upper layers Most wired links are of
  this type
• Reliable, unordered frame service
• Corrupt frames are recovered, frame sending order
  may not be preserved, also duplicate packets
  possible?
• Good for links where the error rate is high such
  as wireless links
• Reliable, ordered frame service
• Corrupt frames are recovered, frames are
  delivered in the order they were sent, no
  duplicate frames possible.
• Very restrictive! X.25 links used this model

5
Steps in Transmission of a Datagram to a Neighbor
Datagram
Datagram
H
Datagram
Y
All bits in D OK?
LL
N
Datagram
H
Detected error
LL
D
EDC
Datagram
H
EDC
Decode Bits from the Link
Encode Bits to the Link
PL
PL
6
General LL Frame Format

• All LL headers must contain a mux/demux key
  (Type)
• Type field in LL header specifies upper layer
  protocol
• IP is one of many upper layer protocols (IPX,
  AppleTalk, ARP, )
• Each LL defines its own protocol type numbering
  for upper layers
• LL header might also contain some control fields
• Does the LL require reliable delivery, flow
  control
• In a broadcast link, header contains the source
  and destination MAC addresses

7
Upper layer protocol numbersin Ethernet

• http//www.cavebear.com/CaveBear/Ethernet/type.htm
  l
• Some Ethernet protocol types
• 0800 DOD Internet Protocol (IP)
• 0806 Address Resolution Protocol (ARP)
• 8037 IPX (Novell Netware)
• 80D5 IBM SNA Services
• 809B EtherTalk (AppleTalk over Ethernet)

8
Maximum Transfer Unit (MTU)

• Each LL has a limit on the max. LL frame size
• This limit is called the Maximum Transfer Unit
  MTU of the link
• MTU is the maximum frame size including the LL
  header, trailer etc.
• If a higher layer protocol such as IP wants to
  send a datagram bigger than the LL MTU, it has to
  break the datagram into smaller pieces
  (fragments), so that each fragment is smaller
  than the MTU
• This is called fragmentation (at the sender) and
  de-fragmentation (at the receiver)

9
Error Detection/Correction

• Errors caused by signal attenuation, electrical
  interference or thermal noise.
• Receiver detects presence of errors, take an
  action
• Possible actions
• Correct bit errors if possible and continue
• Signal sender for retransmission (X.25)
• Drops frame Most common (PPP, Ethernet, )

10
Error Detection/Correction General Idea

• To detect and correct errors, we add extra error
  detection and correction bits (EDC) to the frame

• D Data protected by error checking may
  include header fields
• Error detection not 100 reliable!
• protocol may miss some errors, but rarely
• larger EDC field yields better detection and
  correction

11
Error detection/correction
Error Detection/Correction
Parity Checks
Checksum
Cyclic Redundancy Checks
12
Checksums

• Receiver
• compute checksum of received segment
• check if computed checksum equals checksum field
  value
• NO error detected
• YES no error detected. But maybe errors
  nonetheless?
• Mis-ordered words

• Sender
• treat segment contents as sequence of 16-bit
  integers
• checksum addition (1s complement sum) of
  segment contents

• Why use checksum?
• simple to implement in software
• Typically used by upper layer protocols TCP,
  UDP, IP..
• Found to be enough in ARPANET by experience

13
IP Checksum Algorithm

• unsigned short IPCheckSum(unsigned short buf,
  int count)
• register unsigned long sum 0
• while (count--)
• sum buf
• if (sum 0xFFFF0000)
• / carry occured, so wrap around /
• sum 0xFFFF
• sum
• //end-if
• //end-while
• return unsigned short((sum 0xFFFF))
• //end-IPCheckSum

14
Parity checking

• Single Bit Parity
• Add an extra bit to a 7-bit code to balance the
  of 1s in a byte
• Odd parity sets the 8th bit to 1 to give an odd
  of 1s in the byte
• Even parity sets the 8th bit to 1 to give an
  even of 1s in the byte
• Detect single bit errors

• Two Dimensional Bit Parity
• Computes parity for each bit position across
  each of the bytes contained in the frame
• This results in an extra byte for the entire
  frame in addition to a parity bit for each byte
• Detect and correct single bit errors

parity bit
7 data bits
parity bit
1
0 1 0 1 0 0 1
7 data bits
0
1 1 0 1 0 0 1
1
Data
0 1 0 1 0 0 1
1
1 0 1 1 1 1 0
Even parity example
1
0 0 0 1 1 1 0
Parity Byte
1
0 0 1 0 0 0 0
15
Two-Dimensional Parity Checking
16
Cyclic Redundancy Checks (CRC)

• Idea
• We have a n-bit message M to send
• Treat message bits as coefficients of n-bit
  polynomial (n-1 degree polynomial) M(x)
• Append r bits called the CRC-bits to the end of
  the message such that the new message M of
  length nr-bits is divisible by a generator
  polynomial G(x)
• G(x) is well known and agreed upon in advance
• CRC-16 x16 x15 x2 1 (used in HDLC)
• CRC-CCITT x16 x12 x5 1
• CRC-32 x32 x26 x23 x22 x16 x12 x11
  x10 x8 x7 x5 x4 x2 x 1 (used in
  Ethernet)
• Send M(x), which is divisible by G(x), to the
  receiver

17
Cyclic Redundancy Checks (CRC)

• Given n-bit message M, how to compute M of
  length nr-bits?
• Think of n-bit message M as being represented by
  a n-1 degree polynomial, where the value of each
  bit (0 or 1) is the coefficient for each term in
  the polynomial.
• E.g., the message M(x) 10011010 corresponds to
  the polynomial x7 x4 x3 x
• Consider the generator polynomial G(x) of degree
  k
• Suppose G(x) x3 1. In this case k 3.
• Multiply M(x) by xk to obtain P(x) of degree
  n-1k
• Do polynomial division of P(x) by G(x) to obtain
  remainder R(x)
• Add R(x) to P(x) to obtain M(x)
• Notice that M(x) is divisible by G(x), i.e., the
  remainder is 0

18
Example CRC Computation

• Data
• 101110
• Generator Polynomial
• x3 1 (1001)
• Send
• 101110011

Data
R
19
CRC - Receiver Algorithm

• Assume that the receiver received T(x) of length
  nk-1 bits. What does the receiver do with this
  message?
• Assume T(x) M(x) E(x)
• Divide T(x) by G(x) to obtain the remainder R(x)
• If R(x) is 0, two things are possible
• There is no error in the message, i.e., E(x) 0
• E(x) is divisible by G(x)
• A clever choice of G(x) can make the second case
  extremely rare so that we can safely conclude
  that an R(x) 0 result means that the message is
  error free and non-zero means that an error has
  occurred

20
Forward error correction

• FEC
• Use error correcting codes to repair losses
• Add redundant information which allows receiver
  to correct bit errors
• Requires knowledge at information and coding
  theory and is out of the scope of this course

21
Framing

• Up until now, we have added LL header to the data
  and appended a EDC to the end to form a LL frame
• This frame will be transmitted to the destination
  (receiver)
• Problem How would the receiver determine where
  the frame bits start and where they end as the
  adapter is receiving a sequence of bits from the
  link?
• Called the framing problem
• Idea Add some extra framing information before
  sending the frame

22
Framing

• Several solutions exist to solve the framing
  problem. We will look at a few
• Byte counting Approach DDCMP
• Before sending the frame bits, send of bytes in
  the frame first
• Sentinel Approach - BISYNC, IMP-IMP, HDLC, PPP
• Denote the end (and maybe the beginning) of the
  frame by a special char, called the sentinel char
  or flag
• Q What if the sentinel char appears in the data?
• Byte stuff the control chars
• Bit stuff data sequence
• Physical Layer Coding Violations Ethernet, 802
  LANs
• Make use of physical layer. Use an encoding
  sequence that is not used for encoding data to
  mark the end of a frame
• Requires a dependence between LL PL

23
Byte-Counting Approach DDCMP
8
8
8
16
42
14
Body (variable size)
DDCMP frame format

• Count specifies the size of the frame body
• Problem What if a transmission error corrupts
  the count field?
• Solution
• Collect as much data as the count field, do a
  CRC. If the CRC is bad, then discard the frame
  and wait until the next SYN char to start
  collecting the bytes for the next frame
• Resynchronization can fail if start of frame
  character is inside packets as well BUT will
  eventually sync up
• Not a very popular technique

24
Sentinel Approach PPP
8
16
8
Body (variable size)
PPP frame format

• Each frame starts and ends with the sentinel char
  0x7e
• Uses byte-stuffing to solve the problem of 0x7e
  appearing inside the body
• Sender
• Adds (stuffs) extra lt 01111101gt 0x7d byte
  before each lt01111110gt 0x7e data byte
• 0x7d is also escaped with a 0x7d
• Receiver
• Two 01111101 0x7d bytes in a row discard first
  byte, continue data reception
• 0x7d followed by 0x7e discard first byte,
  continue data reception
• Single 01111110 0x7e flag byte End of frame

25
Byte Stuffing in PPP Example
26
HDLC Bit Stuffing
Body (variable size)
0x7e
0x7e
HDLC frame format

• Each frame starts and ends with the sentinel char
  0x7e (0x01111110)
• When the link is idle, 0x7e is transmitted to
  keep the clocks synchronized
• Uses bit-stuffing to solve the problem of 0x7e
  appearing inside the body
• Sender
• A 0 bit is stuffed after any 5 consecutive 1s,
  i.e., 11111
• Receiver
• After reception of 5 consecutive 1s, examine the
  6th bit
• If 6th bit is 0, remove it (stuffed bit)
• If 6th bit is 1 and 7th bit is 0 (end of frame)
• If 6th bit is 1 and 7th bit is 1 (sender is
  indicating an abort condition)

27
Bit-Stuffing in HDLCExample
28
Physical Layer Coding Violations

• Make use of physical layer encoding scheme.
• Use an encoding sequence that is not used for
  encoding data to mark the end of a frame
• Only applicable to links in which the encoding on
  the physical medium contains some redundancy
• Used by Ethernet and IEEE 802 LANs
• No need for byte/bit stuffing (not even length
  field!)
• More efficient use of link bandwidth
• But, requires a dependence between LL PL
• LL requires a specific PL to run.
• Thats why Ethernet and IEEE LAN specs combine LL
  and PL specs together

29
Recap

• Up until now, we have a NL datagram coming into
  to LL
• LL adds LL header (H)
• Done in software
• LL computes EDC
• Hardware (Ethernet) or Software (PPP)
• LL adds framing info (if necessary)
• Hardware or Software (PPP)
• PL encodes the bits to the link
• Hardware
• This algorithm is enough for a point-to-point
  link.
• What about broadcast links?
• Next