INFO 330 Computer Networking Technology I

1
INFO 330Computer Networking Technology I

• Chapter 5
• The Link Layer LANs
• Glenn Booker

2
The Link Layer

• So, lets see where weve been
• The transport layer provides process to process
  communication
• The network layer provides host to host
  communication
• Now the Link Layer provides the ability to send
  packets across a single link
• So this layer tells how to send a packet/segment/
  datagram from one router/host to another

3
The Link Layer

• There are two types of link layer channels
• Broadcast channels, used in LANs, wireless LANs,
  satellite networks, and HFC cable networks
• Point-to-point communication link, such as
  between two routers or between an ISP and a
  modem
• Well focus on Ethernet and PPP (Point-to-Point
  Protocol)
• Wi-Fi (IEEE 802.11 protocols) is in chapter 6

4
Link Layer Terms

• A node is a router or host at this layer we
  dont care which one were dealing with!
• Any connection between nodes is a link
• The transmitting node puts the datagram in a
  frame, and transmits it into the link
• The receiving node receives the frame, and
  extracts the datagram

5
Link Layer Services

• A link layer protocol moves a datagram over a
  (one, individual, eins, uno) link
• It defines the format of packets (frames)
  exchanged between nodes at each end of the link,
  and the actions the nodes do to send and receive
  these packets
• Over a host-to-host route, links may use several
  different link-layer protocols but one per link
• Typically, one link layer frame contains one
  network layer datagram

6
Link Layer Services

• The link layers actions can also include
• Reliable delivery
• Error detection
• Retransmission
• Flow control
• Random access
• Link layer protocols include PPP, Ethernet, Token
  Ring, Wi-Fi, and some parts of ATM

7
Link Layer Services

• Now elaborate a little on these services
• Framing a datagram into a frame means we have
  data (the datagram) and one or more headers
• Technically, can have header and trailer fields,
  but well generically call both headers
• Header format is defined by the protocol

8
Link Layer Services

• Link Access uses the Medium Access Control (MAC)
  protocol to define how a frame is transmitted
  over a link
• MAC negotiates transmission when many nodes share
  the same link
• Reliable delivery is provided by high error- rate
  links (e.g. wireless) to keep the transport layer
  from retransmitting over the entire route

9
Link Layer Services

• Flow control helps keep the sending node from
  overwhelming the receiving node
• Error detection looks for bit errors, usually
  more elaborately than in the transport and
  network layers
• Error correction some protocols (ATM) can also
  fix errors detected

10
Link Layer Services

• Half vs full duplex with half duplex, a node
  can only send or receive at one time with full
  duplex, it can send and receive at the same time
• Yes, lots of the link layer services are similar
  to transport layer services
• But the link layer only provides them between
  two nodes, whereas the transport layer does
  between hosts

11
Adapters

• Most link layer protocols are implemented in an
  adapter (since were getting really close to the
  physical layer!)
• Adapter network interface card (NIC)
• The adapter is the last connection between a
  host and the physical link to the network
• Error checking occurs in the adapter, oblivious
  to the host
• Only datagrams which come in cleanly are passed
  up the protocol stack to the application

12
Adapters

• The main parts of an adapter are the link
  interface and the bus interface
• The link interface connects to the physical
  network
• The bus interface connects to the parent nodes
  I/O bus (e.g. PCI, PCI-X, Serial ATA, IDE, etc.)
• Not much to it!

13
Error Detection and Correction

• We can detect, and sometimes correct, bit errors
  at the link layer

14
Error Detection and Correction

• We add error-detection and correction (EDC) to
  the data (D) to be sent across the link, in
  addition to other header info (address, sequence
  number, etc.)
• At the other end of the link, the data could be
  changed (D) and the EDC info could be corrupted
  (EDC)
• Telling from D and EDC if the original D was
  corrupted isnt a perfect science!

15
Error Detection and Correction

• Hence there could be undetected bit errors
• The lower the undetected error rate, the larger
  the overhead to add to each frame
• Three main methods for detection
• Parity Checks
• Checksum
• Cyclic Redundancy Check (CRC)

16
Parity Checks

• A simple error detection scheme, parity check
  adds one bit to the data
• That one bit depends on the type of parity
  scheme
• For even parity, the parity bit is chosen so that
  the total number of 1s in the frame is even
• For odd parity, the parity bit is chosen so that
  the total number of 1s in the frame is odd

17
Parity Checks

• If the receiver of an even parity link finds an
  odd number of parity, then there must have been
  some odd number of bit errors (1, 3, 5, )
• Notice that an even number of errors isnt
  detected!
• And yes, it helps if both sides of the link are
  using the same parity rules
• Modems used to set even or odd parity

18
Parity Checks

• A better approach is to break the data into a
  table with i rows and j columns, and define
  parity for each row and column
• In this two-dimensional parity check, there are
  ij1 parity values (bits)
• But by cross-referencing the parity errors,
  exactly which bit(s) were in error can be known,
  and hence fixed!

19
Parity Checks

• If the receiver can detect and fix errors, its
  forward error correction (FEC)
• Commonly used in audio devices to compensate for,
  e.g., scratched CDs
• In a network, this helps avoid retransmission,
  and the associated delays

20
Checksum Methods

• Yup, this is just like the approach we saw
  beforehere we call it an Internet checksum
• Add the digits of the data
• Take the 1s complement of the result thats
  the checksum
• Data checksum 111111111 if not, theres an
  error somewhere
• See RFC 1071

21
Cyclic Redundancy Check

• A Cyclic Redundancy Check (CRC) code is widely
  used in the link layer
• Checksums are easy to calculate in software, so
  theyre ok for the transport and network layers,
  but here we can use hardware to calculate CRC
  codes for us
• The use of CRC codes provides more sophisticated
  error checking

22
Cyclic Redundancy Check

• CRC uses modulo-2 arithmetic, a.k.a. Boolean
  arithmetic
• Its equivalent to XOR (exclusive OR)
• A B (A xor B)
• 0 0 0
• 0 1 1
• 1 0 1
• 1 1 0

23
Cyclic Redundancy Check

• Multiplication by 2k moves the bits left byk
  places
• 1011 23 1011000 (118 6416888)
• So much for the math lesson, so what?
• The CRC code defines the r CRC bits with a
  value of R
• Theres a generator, G, which has some value
  starting with 1, and has r1 bits

24
Cyclic Redundancy Check

• Assume our data has d bits, and is a string
  called D
• The value of R is defined so that D 2r XOR R
  is equal to some exact integer multiple of G
• (D 2r) XOR R nG
• So R remainder D2r / G

25
Cyclic Redundancy Check

• The value of G is typically predefined by IEEE
  standards
• Standard G lengths are 8, 12, 16, and 32 bits
• Hence the corresponding lengths of R are r 7,
  11, 15, and 31 bits

26
Cyclic Redundancy Check

• So how does this mess work?
• Pick a length of G
• Calculate R from the previous slide for each
  data frame, D
• Send the frame
• The receiver divides the dr bits by G
• If the remainder is zero, there are no errors
• If the remainder is not zero, there were errors

27
Cyclic Redundancy Check

• So what? Why all this work?
• Errors tend to occur in bursts not one error
  all by itself
• Using CRC codes allows you to catch up to r
  errors in a single frame
• And errors of more than r in a frame might be
  caught, (1 - 0.5r)100 percent of the time
• And this will catch any number of odd errors
• So thats why we use it a lot at the link layer

28
Multiple Access Protocols

• One type of protocol mentioned earlier is a
  broadcast link
• A node sends a frame to all of the other nodes
• Used by wired, wireless, and satellite networks,
  plus the occasional cocktail party
• This motivates the multiple access problem how
  do we control transmission onto a shared
  broadcast channel

29
Multiple Access Protocols

• Frames can arrive at a node (yes, technically the
  adapter on that node) at the same time, producing
  a collision (both frames on top of each other, a
  mess)
• Dozens of multiple access protocols have been
  defined, but they fall into three types
• Channel partitioning protocols
• Random access protocols
• Taking-turns protocols

30
Multiple Access Protocols

• We want multiple access protocols to provide
• One node can send data at a rate of R bps
• If M nodes want to transmit, each can transmit an
  average of R/M bps
• The protocol should be decentralized, so that a
  single point failure doesnt take down the system
• Its cheap to implement

31
Channel Partitioning Protocols

• Could use FDM or TDM (frequency or time division
  multiplexing) to share a channels bandwidth
  across some number of slots
• Avoids collisions, which is good
• But each slot only gets a fraction of the
  bandwidth, even if no one else is transmitting

32
Channel Partitioning Protocols

• Instead use Code Division Multiple Access (CDMA),
  which assigns codes to each node which sends data
  
• CDMA is also good for avoiding signal jamming,
  hence is used by the military
• Is used widely for wireless protocols

33
Random Access Protocols

• Here each node transmits as though it has the
  full channel bandwidth available
• When a collision occurs, it waits a random amount
  of delay time before retransmitting
• Keep retransmitting until the frame gets through
• There are many protocols of this type, e.g.
• Slotted ALOHA
• ALOHA
• CSMA (of which Ethernet is an example)

34
Slotted ALOHA

• Suppose
• All frames have size L bits
• Time is divided into slots of duration L/R
  seconds ( time to transmit one frame)
• Nodes only transmit at the start of a slot
• Nodes all know when the slots start
• If a collision occurs, the nodes know that before
  the end of the slot occurs
• There is a probability, p, between 0 and 1

35
Slotted ALOHA

• Slotted ALOHA works like this
• When a node needs to transmit a frame, it waits
  until the next slot starts and transmits it
• If theres no collision, the node can transmit
  the next frame if needed
• If there was a collision, the next time a random
  number is greater than p, transmit in that slot
• So if the random value is less than 1-p, wait for
  retransmission

36
Slotted ALOHA

• This takes advantage of the link when only one
  node is active it gets the full rate
• If there are multiple active nodes, some slots
  will be wasted because nobody is transmitting
• The efficiency is the percent of slots where a
  successful transmission occurs
• The efficiency for N active nodes is
  Np(1-p)(N-1)
• Bad part is max efficiency is only 37

37
ALOHA

• What is we ignore the part about transmitting
  only at the start of a slot?
• Transmit when you want to
• If theres a collision, retransmit immediately if
  value is gtp, otherwise wait one slot duration and
  reevaluate retransmitting then
• The icky part is that the efficiency of this is
  only half of Slotted ALOHA the price for
  decentralized control

38
CSMA

• CSMA (Carrier Sense Multiple Access) pays
  attention to whether anyone else is transmitting,
  before a node does so
• Like listening for a break in conversation before
  jumping in, carrier sensing listens for a break
  in link traffic (basic CSMA protocol)
• Collision detection is done by sensing if another
  node starts transmitting while you are (CSMA/CD)

39
CSMA

• There are many variations on CSMA CSMA/CD
• Collisions can occur because of the time needed
  for transmitting frames the channel propagation
  delay

40
Taking-turns Protocols

• The ALOHA and CSMA protocols both take advantage
  of full bandwidth when available, but neither is
  good at assuring fair share of throughput when
  multiple nodes are active
• To fix the latter, taking-turns protocols have
  been made hundreds of them!
• Well focus on two major kinds
• Polling protocols
• Token-passing protocols

41
Polling Protocols

• Polling protocols make one node a master node
• The master node polls each node in turn, and
  tells each node it can send some number of frames
• This eliminates collisions and empty slots
• But it adds a polling delay to notify each node
  its turn is up, and delays to check nodes which
  are inactive
• And its really bad if the master node dies!

42
Token-passing Protocols

• Token-passing protocols have no master node, but
  instead pass a small token frame among the nodes
  in a fixed order
• Each node holds the token only if they have
  frames to transmit, up to some max number
• Then keep passing the token
• Failure of ANY node crashes the network!
• Or if the token isnt released, theres trouble
• FDDI and yes, Token Ring, are examples

43
Local Area Networks (LANs)

• Local Area Networks use multiple access protocols
  extensively
• Ethernet is the most common random access
  protocol
• Token Ring had a slight speed advantage, so it
  was popular in the late 1980s
• A node sends a frame around the network, and
  its read by the recipient node
• The sender removes it from the network

44
Local Area Networks (LANs)

• FDDI (Fiber Distributed Data Interface) was
  designed for larger LANs, specifically
  Metropolitan Area Networks (MANs)
• Under FDDI, the destination node removes the
  frame from the network
• Hence it isnt a pure broadcast channel, since
  nodes downstream will never get the frame

45
Link Layer Addressing

• Cmon, we havent had an address format in at
  least two or three days
• Here well go over MAC, ARP, and DHCP
• As stated earlier, the adapter is the real
  location of a link layer address
• The MAC address (a.k.a. LAN address or physical
  address) is the link layer address of an adapter

46
MAC Address

• A MAC address usually has 6 bytes, so there are
  248 MAC addresses
• 248 281,474,976,710,656 in case you wondered
• Each byte is expressed as two hexadecimal numbers
  (0-9, A-F for 10-15)
• 01904B5F3113
• The IEEE makes sure each MAC address is unique

47
MAC Address

• MAC addresses have no other structure, and never
  change for a given adapter
• Like the IP address, the MAC address is used to
  verify that the destination host (adapter) has
  been reached
• The MAC broadcast address is all Fs, analogous
  to the 255.255.255.255 IP address
• FFFFFFFFFFFF

48
Address Resolution Protocol

• The Address Resolution Protocol (ARP) (no, not
  AARP) translates between IP addresses and MAC
  addresses
• RFC 826, for the curious
• ARP only works within the local subnet
• Unlike DNS, which resolves addresses anywhere
• Each node (host / router) maintains an ARP table
  to map IP addresses MAC addresses

49
Address Resolution Protocol

• ARP also includes a time-to-live, which is the
  time before that entry is deleted
• Typically starts at 20 minutes and counts down
• A special ARP packet is broadcast to all nodes on
  the subnet to resolve an unknown MAC address
• ARP has query and response packets, both with the
  same format

50
Address Resolution Protocol

• The query is sent in a broadcast frame, but the
  response is sent in a standard frame
• ARP builds itself if it gets an unknown
  address, it works to find the information
• If a node is deleted from the network, its ARP
  entries get removed eventually too

51
ARP Off Subnet

• To send a frame outside of the local subnet,
  first have to use the MAC address of the
  interface leading out of the subnet
• Then the frame goes through a router to the
  correct subnet, where the interface on that
  subnets side can resolve the correct MAC using
  ARP

52
ARP Off Subnet

• A creates datagram with source A, destination B
• A uses ARP to get Rs MAC address for
  111.111.111.110
• A creates link-layer frame with R's MAC address
  as dest, frame contains A-to-B IP datagram
• As adapter sends frame
• Rs adapter receives frame
• R removes IP datagram from Ethernet frame, sees
  its destined to B
• R uses ARP to get Bs MAC address
• R creates frame containing A-to-B IP datagram
  sends to B

53
Ethernet

• Ethernet has been king of wired LANs since the
  late 1970s why?
• 1) it was the first high speed protocol
• 2) its cheap
• 3) it has had speed increases to stay competitive
• The original Ethernet (thick and thin
  Ethernet) used a bus topology

2.94 Mbps in 1973!
54
Ethernet

• But now a hub or switch is used at the center of
  a star topology

55
Ethernet Frame Structure

• Ethernet frames use this structure
• The Preamble is 8 bytes, the first seven of which
  are all 10101010, and the 8th is 10101011
• Used to synchronize the clocks between sender and
  receiver, since many possible speeds could be
  used (10 Mbps to 1000 Mbps)

56
Ethernet Frame Structure

• The Destination Address is the 6-byte MAC
  address of the destination
• The Source Address is the senders MAC
• The Type field is 2 bytes to explain the network
  protocol which created the frame (IP, IPX,
  AppleTalk, etc.)

57
Ethernet Frame Structure

• The Data field is 46 to 1500 bytes for the IP
  datagram, in our case
• Use stuffing to pad the Data to 46 B if needed
• 1500 B is the max transfer unit (MTU) for
  Ethernet
• Finally, the CRC field is a 4 Byte CRC code
  discussed earlier to detect bit errors in the
  frame
• So the Ethernet frame has 26 B of headers plus
  the data field psst! WAKE UP!

58
Ethernet

• Ethernet is connectionless service, like IP and
  UDP theres no handshake at this layer
• Therefore its service is unreliable
• The CRC check is used, but failed frames are
  merely discarded
• A lost frame here means a lost (or incomplete)
  segment at the UDP layer
• Ethernet is blissfully unaware if a frame is new,
  or retransmitted, or even related to any other
  frames

59
Ethernet and CSMA/CD

• If a hub is used, Ethernet broadcasts to all
  nodes (adapters) on the LAN
• Ethernet uses CSMA/CD
• No slots, just start broadcast when ready
• Use carrier sensing to know when NOT to
  broadcast
• Stop transmitting when a collision is detected
• Before retransmitting, wait a short random time

60
Ethernet and CSMA/CD

• Efficiency can reach 100 in a LAN
• It senses a collision, or the lack of traffic by
  monitoring voltage levels on the link
• Pause for an open line is 96 bit times, or 9.6
  microsec at 10 Mbps
• If collision is detected, a 48-bit jam signal is
  transmitted instead of the frame, to all adapters
• Delay for the nth collision is 512K bit times
• K is random from 0,1,2,,(2m 1) where
  mmin(n,10)

61
Ethernet and CSMA/CD

• Notice that the more collisions are noted, the
  longer the possible delay time
• Called an exponential backoff
• Ethernet efficiency is messy to calculate, but
  comes to Efficiency 1 / (1 5tprop/ttrans)
• For small propagation time and/or large
  transmission time, this is about 1

62
Ethernet and LANs

• Ethernet is used for most wired LANs
• 100BaseT and 1000BaseT are common (100 Mbps and
  1000 Mbps, respectively)
• 10 Gigabit Ethernet exists
• A hub is frequently the center of a simple star
  network
• Hubs operate only on physical layer

63
Ethernet and LANs

• Thin Ethernet cable is still available the
  connectors are called male or female BNC
• Image from CompUsa

64
Hubs

• Hub are the village idiot of networking hardware
  ok, maybe a handy village idiot
• When a bit arrives on any of its adapters, it
  copies it, amplifies it a little, and retransmits
  it on all of the other adapters
• They typically have 4-24 adapters, or ports
• Cost is way under 100 for most hubs
• They do nothing for CSMA/CD

65
Hubs

• An adapter may malfunction and keep transmitting
  (a jabbering adapter), in which case the hub
  should detect the problem and shut off that
  adapter
• Fancy hubs can collect and report usage data,
  collision rates, frame sizes, etc
• Max of 100 meters between hub and hosts for
  twisted pair wire more for optical cables
• The T in 100BaseT means twisted

66
Hub Hierarchy

• Hubs can be connected in a multi-tier hierarchy
  so that different parts of a building, or
  different departments, etc. can share resources

67
Hub Hierarchy

• The backbone hub has three LAN segments attached,
  each with its own hub
• This extends the max distance covered
• But these are all part of the same collision
  domain
• All segments have to share same Ethernet speed
• Limits throughput across entire network

68
Link-layer Switches

• Switches operate on the link layer
• Incoming Ethernet frames are examined for the
  layer-2 (link layer) destination (e.g. MAC
  address)
• It then forwards the frame to the adapter leading
  to that destination (not all of the adapters)
• If the backbone hub on slide 68 were replaced by
  a switch, then each LAN segment is now its own
  collision domain

69
Link-layer Switches

• Switches can handle multiple network speeds
• Some segments at 10 Mbps, others at 100 Mbps,
  etc.
• They still allow communication across the
  segments
• They can be combined into any size network
• They operate in full duplex (transmit and receive
  at once) and provide, um, switching

70
Switch Filtering Forwarding

• Filtering is when a switch can decide to forward
  a frame or just drop it
• Forwarding is deciding which interface a frame
  needs to go out on, and directing it there
• A switch table is used for both filtering and
  forwarding

71
Switch Table

• A switch table has the MAC address of each node,
  the corresponding interface number to get to
  that node, and the time the entry was made
• When a frame comes in looking for a given MAC
  address
• If the address is from the same interface it came
  in on, do nothing (the frame is internal to that
  segment) this is filtering the frame

72
Switch Table

• If the address needs to go to another interface,
  send it there
• If the address doesnt exist, see next slide ?
• Recall hubs transmit without concern for existing
  traffic
• A switch uses CSMA/CD to tell when to transmit,
  but its interfaces are not adapters (they have no
  MAC addresses)

73
Switch Learning

• The switch table is built automatically they
  are self-learning
• The switch table is empty to start
• If a frame arrives with a MAC destination not in
  the table, send it to all other interfaces
• Each time a frame is received, record the
  interface and address from which it came, and
  the current time
• If the aging time expires, remove that address
  from the table

74
Switches

• Switches are plug-and-play devices, because they
  configure the switch table automatically
• One can have dedicated access to a switch, with
  separate connections for transmitting and
  receiving data
• This makes collisions impossible for those hosts
• Dedicated access means point-to-point connections
  can be used no multiple access protocol needed!

75
Switches

• Switches can help a network by
• Eliminate collisions, if there are no hubs in
  the network
• Have links at different speeds (but all the same
  protocol)
• Shut off misbehaving adapters

76
Cut-through Switching

• Many switches use store-and-forward packet
  switching, just like routers
• Some use cut-through switching
• A frame is transmitted before it is fully stored,
  as soon as the output link is available
• This can eliminate the store-and-forward delay,
  L/R seconds
• Not a huge savings in most cases

77
Switches vs Routers

• Switches are store-and-forward packet switches
  that use the layer 2 address
• Routers use the layer 3 address
• Switches can only use the spanning tree
  structure
• Routers can use any structure
• Switches are plug-and-play
• Routers need to be maintained manually, e.g.
  their IP addresses

78
Switches vs Routers

• A large switched-only network would need large
  ARP tables, and provides no protection against
  broadcast storms (an errant host transmitting
  endlessly)
• Processing time for a switch is typically less
  than for a router
• Bottom line need to use routers to control
  larger networks, or when its intelligence is
  helpful, use a switch whenever possible

79
Hubs vs. Switches vs. Routers
80
PPPPPPPPPPPPPPPPPPPPPPPP

• The Point-to-Point Protocol (PPP) is the main
  protocol used to connect between an ISP and a
  customer
• Related, but not addressed here, is the
  High-level Data Link Control (HDLC) protocol
• PPP could be used over many types of connection
  a dial-up modem, X.25, SONET (synchronous optical
  network), ISDN, etc.

81
PPP

• PPP requires that
• A network packet be put in a frame so that the
  receiver will be able to tell where the packet
  and frame start and stop
• There are no limits on the format of headers or
  data within the packet
• It support many network protocols (IP, DECnet,
  etc.)
• It must run over many kinds of links (serial,
  parallel, synchronous, asynchronous, any speed,
  electrical or optical)

82
PPP

• Must be able to detect bit errors
• Must detect link failure (dead link)
• Must provide a way for the network layer protocol
  to learn or configure the network address
• And keep it simple!
• Be grateful other features were NOT included
• Error correction, Flow control, Sequencing
  (frames get there in order)
• Multipoint links (many receivers of a single
  message)

83
PPP Data Framing

• The PPP data frame steals a little from HDLC
• A 1-byte Flag of 01111110 starts and ends each
  frame
• A 1-byte Address of 11111111 is next
• A 1-byte Control field consists of 00000011
• Pretty boring, huh?
• Then a 1-2 byte Protocol field tells what network
  layer protocol will be used (hex 21IP,
  29AppleTalk, 27DECnet)

84
PPP Data Framing

• Then comes the Information being transmitted
• The default max size is 1500 bytes, but that can
  be changed
• Then a 2 or 4 byte Checksum using the format
  defined by HDLC for a CRC code
• Then the other Flag field ends the frame
• So the headers and trailers total 7 to 10 B for
  PPP frames

85
Byte Stuffing

• So what if the data includes the Flag value?
• Add a control escape byte in front of it, to tell
  theres a non-Flag sequence of 01111110
• So the extra byte 01111101 is stuffed in front
  of any actual data bytes which happen to be
  01111110

86
Other PPP Protocols

• PPP uses link-control protocol (LCP) to
  configure the link, much like TCPs handshake
• Once the link is open, PPP works with the
  network-control protocols (network layer
  protocols to control the connection)
• E.g. there are IP control protocols to manage a
  PPP connection

Yes, thats redundant
87
Link Evolution

• A link started as a wire connecting nodes
• But a link can be electromagnetic too
• And a link can be a channel used to communicate
  with many nodes
• At the link layer, we dont typically know or
  care if our link is part of a home network or a
  global WAN we just want our little frames to
  get to the other node

(Skip the ATM discussion in sections 5.8.1 and
5.8.2)
88
Summary

• The link layer gets us from one node (host or
  router) to another
• All link layer protocols take network layer
  datagrams and put them in frames to be sent over
  the physical (though not always solid) medium of
  the link
• A point-to-point link has one sender and one
  receiver multiple access links can have many of
  both