Computer Networks

1
Computer Networks

• Chapter 4
• Medium Access Sublayer
• Prof. Jerry Breecher
• CSCI 280
• Spring 2002

2
The Weeks Ahead
Date Lecture Lab/Project Jan 28 30 Medium
Access Layer Jan 31 Project 1
Checkin Feb 4 - 6 Medium Access Layer Feb
7 Project 1 Checkin Feb 11 Exam 1 Feb
13 The Data Link Layer Feb 14 Project 1
DUE
3
Chapter Overview

• Here we want to know how to handle broadcast
  networks. As compared to point to point
  networks, a major issue is handling arbitration
  when there is competition for the network.
• This is the bottom sublayer of the Data Link
  Layer. This Chapter is especially relevant for
  LANs.
• 4.1 The Channel Allocation Problem
• How to allocate a single channel among multiple
  users.
• 4.2 Multiple Access Protocols
• How to handle contention for the use of a
  channel.
• 4.3 IEEE Standards for LANs
• How do the protocols of the last sections apply
  to real systems. Here we talk about the actual
  standards in use.
• 4.4 Bridges
• Ways of connecting networks together.
• 4.5 High Speed LANs
• Directions in high speed networks.

4
CHANNEL ALLOCATION PROBLEM
Overview
Or, how to allocate a single channel among
competing users.

• 4.1 The Channel Allocation Problem
• 4.2 Multiple Access Protocols
• 4.3 IEEE Standards for LANs
• 4.4 Bridges
• 4.5 High Speed LANs

5
CHANNEL ALLOCATION PROBLEM
STATIC CHANNEL ALLOCATION IN LANs AND MANS
The traditional (phone company) way of allocating
a single channel is Frequency Division
Multiplexing. (See Figure) FDM works fine for
limited and fixed number of users. Inefficient
to divide into fixed number of chunks. May not
all be used, or may need more. Doesn't handle
burstiness. T mean time delay C capacity
(bps) l arrival rate 1/ m mean length
1 T ---------- mC - l
6
CHANNEL ALLOCATION PROBLEM
STATIC CHANNEL ALLOCATION IN LANs AND MANS
Now divide this channel into N subchannels, each
with capacity C/N. Input rate on each of the N
channels is A/N. Now
1 N T(fdm) ----------------- ------------
NT m(C/N) - l /N mC - l
7
CHANNEL ALLOCATION PROBLEM
DYNAMIC CHANNEL ALLOCATION
Possible underlying assumptions include Station
Model - Assumes that each of N "stations"
(packet generators)independently produce frames.
The probability of producing a packet in the
interval dt is ?dt where ? is the constant
arrival rate. That station generates no new
frame until that previous one is
transmitted. Single Channel Assumption
- There's only one channel all stations are
equivalent and can send and receive on that
channel. Collision Assumption - If two frames
overlap in any way time-wise, then that's a
collision. Any collision is an error, and both
frames must be retransmitted. Collisions are the
only possible error.
8
CHANNEL ALLOCATION PROBLEM
DYNAMIC CHANNEL ALLOCATION
Continuous Time - There's no "big clock in the
sky" governing transmission. Time is not in
discrete chunks. Slotted Time
- Alternatively, frame transmissions always
begin at the start of a time slot. Any station
can transmit in any slot (with a possible
collision.) Carrier Sense - Stations can tell a
channel is busy before they try it. NOTE - this
doesn't stop collisions. LANs have this,
satellite networks don't.
9
Multiple Access Protocols
Overview
How to handle contention for the use of a channel.

• 4.1 The Channel Allocation Problem
• 4.2 Multiple Access Protocols
• 4.3 IEEE Standards for LANs
• 4.4 Bridges
• 4.5 High Speed LANs

10
Multiple Access Protocols
MULTIPLE ACCESS PROTOCOLS
Collisions work well for low utilization (they're
not likely to happen.) Arbitration, which we'll
talk about later, works better at high
utilization. ALOHA Developed in Hawaii in the
1970s. PURE ALOHA Every station transmits
whenever it wants to. Colliding frames are
destroyed. The sender knows if its frame got
destroyed, and if so waits a random time and then
retransmits. 1. ANY overlap is a collision. 2.
Best efficiency if frames are same size. 3. BUT,
what is the efficiency? Talk about stochastic.
11
Multiple Access Protocols
MULTIPLE ACCESS PROTOCOLS

• S frames to be transmitted. In units of frames
  per frame time so that 0 the meaning of frame time as used here??)
• G S frames retransmitted due to previous
  collisions.
• P0 probability that a frame does NOT suffer
  collision.
• S P0 x G
• Use the Figure to determine collision
  vulnerability.

Pure Aloha Continued
12
Multiple Access Protocols
MULTIPLE ACCESS PROTOCOLS
Probability that k frames are generated during a
given frame time (Poisson distribution)
G k e-G Prk --------------
k! Probability of no traffic initiated during
the vulnerable period P0 e-2G
so Throughput per frame time is S G e
-2G See Figure 4.3
Pure Aloha Continued
13
Multiple Access Protocols
MULTIPLE ACCESS PROTOCOLS
SLOTTED ALOHA Doubles efficiency by dividing
time into "ticks". Sends occur only at the start
of a tick. Vulnerable period is 1/2 of pure
Aloha case, so S G e-G See throughput
on the last page. Best throughput is at G 1
when S 0.37 empty slots Pr0
0.37 collisions G - S 0.26
14
Multiple Access Protocols
CARRIER SENSE MULTIPLE ACCESS PROTOCOLS
This is where the sender listens before ejecting
something on the wire. Collision occurs when a
station hears something other than what it
sent. PERSISTENT AND NONPERSISTENT
CSMA 1-persistent CSMA Station listens. If
channel idle, it transmits. If collision, wait a
random time and try again. If channel busy, wait
until idle. If station wants to send AND
channel idle then do send. Success
here depends on transmission time - how long
after the channel is sensed as idle will it stay
idle (there might in fact be someone else's
request on the way.) Nonpersistent CSMA
(equivalent to 0-persistent CSMA) Same as above
EXCEPT, when channel is found to be busy, don't
keep monitoring to find THE instant when it
becomes free. Instead, wait a random time and
then sense again. Leads to 1) better
utilization and 2) longer delays than 1 -
persistent. (why?)
15
Multiple Access Protocols
CARRIER SENSE MULTIPLE ACCESS PROTOCOLS
Nonpersistent CSMA (equivalent to 0-persistent
CSMA) Same as above EXCEPT, when channel is
found to be busy, don't keep monitoring to find
THE instant when it becomes free. Instead, wait
a random time and then sense again. Leads to 1)
better utilization and 2) longer delays than 1 -
persistent. (why?)
p-persistent CSMA For slotted channels. If
ready to send AND channel idle
then send with probability p, and with
probability q 1 - p defers to the next
slot. Interpret the chart for these shown in the
Figure.
16
Multiple Access Protocols
CARRIER SENSE MULTIPLE ACCESS PROTOCOLS
CSMA WITH COLLISIONS DETECTION CSMA/CD - used
with LANs. When a station detects a collision,
it stops sending, even if in mid-frame. Waits a
random time and then tries again. What is
contention interval -- how long must station wait
after it sends until it knows it got control of
the channel? It's twice the time to travel to
the furthest station.
17
Multiple Access Protocols
COLLISION-FREE PROTOCOLS
How long is a packet (or how long a wire is
needed to contain a packet) of length 1500 bytes
on a 100 Mbps ethernet? As cables become longer
and faster, the above methods become less
efficient. So, . . . . Bit map protocol -
A "contention slot", subdivided into bits,
allows each station to announce that it wants to
send. After the announcement, all stations can
send in priority order, and there will be no
fighting over the channel. Called "reservation
protocol". What are pros and cons of this
method? Analyze at low and high loads. Binary
Countdown - In the contention slot, each station
places its ID. They all get ord on top of each
other. A particular station knows if it won
because no wanting-to-send station had a higher
number than it did in the slot. ( For instance,
101101 OR 110011 The 101101 station
knows it lost by the time it sends its second bit
- it sees a 1 on the wire when it just sent out
a 0, so it knows the game is up.
18
Multiple Access Protocols
LIMITED-CONTENTION PROTOCOLS
Collision techniques work well for low
utilization (they're not likely to happen.)
Arbitration, which we'll talk about later, works
better at high utilization. This method provides
best of these techniques. Divide the stations up
into groups. Stipulate that only members of
group 0 can arbitrate for slot 0, members of
group 1 for slot 1, etc. Works because it cuts
down on the contention felt by any particular
station. Want a method that will have many
members per group at low contention, and few (or
one) member at high contention. Can use a binary
search to do this.
19
Multiple Access Protocols
Wireless LAN Protocols IEEE 802.11

• Wide Range of uses
• Infrared signals within a building
• Mobile computing
• Network of low flying satellites
• Physical Properties
• The spec allows running over three possible media
  
• Radio using frequency hopping
• Radio using direct sequencing
• Infrared over short distances ( 10 meters )
• The explanation of this requires a detour into
  spread spectrum radio

20
Multiple Access Protocols
Wireless LAN Protocols IEEE 802.11

• Spread Spectrum Radio
• Purpose is to spread signal over a wider
  frequency range so as to minimize interference (
  military uses this for anti-jamming ).
• So the signal can be in a very noisy environment
  and still get through.
• Frequency Hopping
• Transmit the signal over a pseudo-random sequence
  of frequencies. 1st one frequency, then a
  second, etc. The sender and receiver are using
  the same random number generator so they can stay
  in sync.
• The spec calls for using 79 different 1 MHz wide
  bandwidths.
• Direct Sequence
• Each bit of data is replaced by multiple bits in
  the signal.
• Transmitter sends the exclusive or of the data,
  PLUS n random bits.
• Again, both the sender and receiver know the
  random sequence.

21
Multiple Access Protocols
Wireless LAN Protocols IEEE 802.11
Direct Sequence Example
A.
B.
C.
A. Data Stream 1 0 1 0
B. Random Sequence 0100101101011001
C. XOR of the two 1011101110101001
22
Multiple Access Protocols
Wireless LAN Protocols IEEE 802.11

• Collision Avoidance
• Similar to Ethernet, but not quite the same.
• But more complicated because all nodes dont see
  each other.
• Example of the problem
• A and C send a signal to B.
• But A and C arent aware of each others signals.
• Signals collide at B.
• But A and C dont know they collided so dont
• go into collision avoidance.
• A and C are hidden nodes
• Example of another problem
• B is sending to A.
• C can hear this signal from B
• C assumes it can NOT transmit
• But C could in fact transmit to D
• This is called the exposed node problem

C can see this range
A.
D.
B.
C.
23
Multiple Access Protocols
Wireless LAN Protocols IEEE 802.11

• Collision Avoidance
• Problem is solved by using a protocol Multiple
  Access with Collision Avoidance ( MACA).
• Sender and receiver exchange control frames so
  the transaction goes like this
• Sender (A) does Request to Send (RTS) to
  receiver(B)
• B sends back Clear to Send (CTS)
• A sends packet.
• B sends an ACK after receiving the frame.
• Logic
• If a node hears the CTS, it knows it is
• near the receiver - so dont transmit.
• If a node hears the RTS but not the CTS,
• its not near the receiver so it can
• start its own transaction.
• RTS and CTS contain length of packet
• to be sent so others know how long to wait.

C can see this range
A.
D.
B.
C.
24
Multiple Access Protocols
Wireless LAN Protocols IEEE 802.11
Distribution System Some nodes roam (A - H in
the figure.) Some nodes are wired together --
Access Points (AP in the figure). These APs
are called the distribution system.
Distribution System
The three regions shown are like cells in a cell
phone system. Two nodes ( A B) could
communicate with each other directly, but in
practice they go thru the APs. The path from A
to E is A AP-1 AP-3 E.
AP-1
AP-3
AP-2
A.
B.
F.
H.
G.
C.
D.
E.
Nodes
25
Multiple Access Protocols
Wireless LAN Protocols IEEE 802.11
Distribution System Protocol for how a node
finds an access point
1. The node sends a Probe frame. 2. All APs
within range respond with a Probe Response
frame. 3. The nodes selects one of the APs
sends that AP an Association Request frame. 4.
That AP replies with an Association Response
frame. APs also send a Beacon frame that
advertises they are available.
Distribution System
AP-1
AP-3
AP-2
A.
B.
F.
H.
G.
C.
D.
E.
26
Multiple Access Protocols
Wireless LAN Protocols IEEE 802.11
Frame Format Contains the following
fields Control - is the frame carrying data or
is it RTS or CTS or is it forwarding
data. Payload - up to 2312 bytes of data CRC -
checksum of the packet. Addr(I) - Its possible
that the packet needs to be sent across the
distribution system in which case we keep track
of the original sender and the original receiver,
but we also want to know intermediate senders and
receivers.
16 16 48 48 48 16
48 0-2313 32
Control Duration Addr1 Addr2 Addr3 SeqCtrl
Addr4 Payload CRC
27
IEEE Standard 802 For LANs and MANs
Overview
How do the protocols of the last sections apply
to real systems. Here we talk about the actual
standards in use. 802.2 Describes the upper
part of the data link layer, the LLC (Logical
Link Control). Descriptions of the physical and
lower part of the DLL are 802.3 Is CSMA/CS
LAN 802.4 Is Token Bus 802.5 Token Ring

• 4.1 The Channel Allocation Problem
• 4.2 Multiple Access Protocols
• 4.3 IEEE Standards for LANs
• 4.4 Bridges
• 4.5 High Speed LANs

28
IEEE Standard 802
IEEE STANDARD 802.3 ETHERNET

• This is a 1-persistent CSMA/CD LAN. Originated
  in Aloha.
• WIRES

29
IEEE Standard 802
IEEE STANDARD 802.3 ETHERNET

• Wiring

Repeaters - Multiple cables can be connected.
From software point, a repeater is transparent.
30
IEEE Standard 802
Manchester Encoding

• Life would be easy if
• binary 0 0 volts
• binary 1 5 volts
• But there's no way to distinguish a 0 from
  nothing-happening. Need to know when is middle
  of bit WITHOUT a clock.

31
IEEE Standard 802
802.3 MAC SUBLAYER PROTOCOL

• Preamble 7 bytes of 10101010
• Start 1 byte of 10101011
• Dest 6 bytes of mac address
• multicast sending to a group of
  stations.
• broadcast (dest. all 1's) to all stations
  on network
• Source 6 bytes of mac address
• Length number of bytes of data
• Data comes down from network layer
• Pad ensures 64 bytes from dest addr
  thru checksum.
• The pad ensures transmission takes enough time so
  it's still being sent when the first bit reaches
  the destination. The frame needs to still be
  going out when the noise burst from another
  stations collision detection gets back to the
  sender.
• checksum 4 bytes of CRC.

Packet Definition
32
IEEE Standard 802
802.3 MAC SUBLAYER PROTOCOL
Packet Definition
Why you need minimum Packet Size.
33
IEEE Standard 802
802.3 MAC SUBLAYER PROTOCOL

• BINARY EXPONENTIAL BACKOFF ALGORITHM
• After a collision, station waits 0 or 1 slot. If
  it collides again while doing this send, it picks
  a time of 0,1,2,3 slots. If again it collides
  the wait is 0 to 23 -1 times. Max time is 210
  -1 (or equal to 10 collisions.) After 16
  collisions, an error is reported.
• Slot is determined by the worst case times 500
  meters X 4 repeaters 512 bit times 51.2
  microseconds.
• 500 m ? Rep ? 500 m ? Rep ? 500 m ? Rep ? 500 m ?
  Rep ? 500 m ?
• Algorithm adapts to number of stations.

34
IEEE Standard 802
802.3 PERFORMANCE

• Note that channel efficiency depends on
• F -- frame length,
• B -- network bandwidth,
• L -- cable length
• c -- speed of signal propagation
• e -- optimal number of contention slots per
  frame. (512 bits 64 bytes means a 64 byte
  frame has value 1.) BUT, this is not the
  optimal value.
• 1
• channel efficiency ---------------------
• 1 2 B L e /
  c F
• Note Efforts focus on improving both B and L,
  both of which will decrease efficiency.
• Note on traffic patterns arrivals are not
  Poisson, but self similar. This means that
  fluctuations occur on any observation scale (kind
  of like fractals.)

35
IEEE Standard 802
Switched 802.3 LANs

• Uses 10Base-T to each of the hosts. And a high
  speed backplane between the connectors. Works
  because the assumption is that many requests can
  be routed within the switch. Relieves congestion
  on the hub.
• Routing -
• Local (on-switch) destinations are sent there
  directly. Off-switch are sent to the backplane.

Collision Detection - The connections on the
switch form their own LAN and do collision
handling as we've just seen. The switch buffers
the transmission and ensures no collisions occur.
36
IEEE Standard 802
IEEE STANDARD 802.4 TOKEN BUS

• Need a mechanism to handle real-time,
  deterministic requirements. 802.3 could contend
  forever and this is often not acceptable.
• A ring, with stations taking turns is
  deterministic. Uses logical ring on linear
  cable.
• Mechanism -
• o All stations numbered station knows of its
  neighbors.
• o A token, required in order to send, is
  initialized by the highest number station.
• o A station, receiving the token, does a send if
  it has a request, then sends the token to its
  logical (not necessarily physical) neighbor.
• Activation -
• o Stations can come and go on the bus, without
  breaking mechanism.
• Cabling -
• o Uses 75 ohm coax. Speeds are 1, 5, 10 Mbps.

37
IEEE Standard 802
IEEE STANDARD 802.4 TOKEN BUS

• TOKEN BUS MAC SUBLAYER PROTOCOL
• Station has 4 possible priorities, 0, 2, 4, 6
  station maintains 4 queues for requests.
• Within each station,
• Token comes first to priority 6 queue. Sends
  occur until nothing to send OR timer expires.
• Token goes next to priority 4 queue. Sends occur
  until nothing to send OR timer expires.
• And so on . . . .
• Proper setting of the various timers ensures that
  high priority requests happen first.

38
IEEE Standard 802
IEEE STANDARD 802.4 TOKEN BUS

• The frame format. Fields are
• Preamble - used to synchronize receiver clock.
• Start/End Delimiter - contains a non-data
  (illegal) Manchester Encoding.
• Frame control - shows control or data. shows
  priority of data packets. flag requiring ACK
  from receiver. shows type of control frame (more
  later).
• Destination Address - (same as 802.3) - usually 6
  bytes.
• Source Address - (same as 802.3) - usually 6
  bytes.
• Data - BIG - 8182 or 8174 bytes (note
  no length field - why not?)
• Checksum - (Same as 802.3)

39
IEEE Standard 802
TOKEN BUS LOGICAL RING MAINTENANCE

• control frames for ring maintenance.
• SOLICIT_SUCCESSOR -
• Gives senders address and the current
  successor's address. Stations not in the ring,
  with address between these two are invited to bid
  to be inserted.
• No response within given time go on as
  before.
• One response newcomer is inserted becomes
  new successor.
• Two or more responses answers collide so
  garbled.

40
IEEE Standard 802
TOKEN BUS LOGICAL RING MAINTENANCE

• control frames for ring maintenance.
• RESOLVE_CONTENTION -
• Causes responding stations to NOT immediately try
  to be successors, but use binary countdown by 0,
  1, 2, or 3 slots. Mechanism also ensures that
  traffic isn't slowed down by solicitation.(limited
  to less frantic times.)
• SET_SUCCESSOR -
• Used by a leaving station. Sent to the
  predecessor to say the leaver's successor is now
  the predecessor's successor.
• WHO_FOLLOWS -
• The token sender listens to make sure the
  successor got and then passed on the token. If
  doesn't happen, it sends a WHO_FOLLOWS and failed
  station's successor sends a SET_SUCCESSOR to the
  failed one's predecessor.
• SOLICIT_SUCCESSOR_2 -
• The token sender can't find the successor and
  there's no response from WHO_FOLLOWS This
  causes ALL stations to once again bid for a place
  in the ring - this is like starting from scratch.
• CLAIM_TOKEN -
• If the token holder crashes, then nothing appears
  on the ring. All station's timers go off and the
  contention algorithm determines who gets to
  generate the token.

41
IEEE Standard 802
IEEE STANDARD 802.5 TOKEN RING

• Not broadcast but point to point.
• All digital rather than analog (such as used by
  802.3 for collision detection.)
• Chosen by IBM for its LAN included by IEEE as
  Token Ring.
• Calculate the number of bits on the ring at any
  one time
• At R Mbps, a bit is emitted every 1/R
  microseconds (usecs).
• At a speed of 200 m/usec, each bit occupies
  200/R meters of the ring. So a 1 Mbps ring,
  with circumference 1000 meters has only 5 bits on
  it at any one time.
• In addition, there's a 1 bit delay at each
  station. (Data bit can be modified before being
  forwarded.)

Token is 3 bytes. Must be sufficient delay on
the ring so that the whole token is there. Why??
Stations may be powered down, etc. - no
guarantee that stations are adding delay. So may
need to add artificial delay.
42
IEEE Standard 802
IEEE STANDARD 802.5 TOKEN RING

• Arbitration -
• Must hold the token in order to transmit.
• Listen mode -
• Input just copied to output.
• Transmit mode -
• Seize the token and put own data on ring. As
  sender's data comes back around, it removes data.
  At end of transmission, stick token back on.
  Receiver can ACK receipt by flipping a bit on end
  of packet.
• Efficiency is excellent At high usage, with
  many stations transmitting, they get token one
  after the other.

43
IEEE Standard 802
IEEE STANDARD 802.5 TOKEN RING

• Wires -
• Shielded twisted pair/ 1 or 4 Mbps.
• Differential Manchester encoding.
• Reliability -- Star Shaped Ring --

44
IEEE Standard 802
TOKEN RING MAC SUBLAYER PROTOCOL

• Frame Structure Components -
• SD, ED Delimiters - have illegal encoding so
  not confused as data.
• AC Access control, containing bits for
• The token bit - flip this bit and its a data
  preamble
• Monitor bit,
• Priority bits,
• Reservation bits
• Frame control Provides numerous control options.
• Source/Destination addresses/checksum
• same as 802.3 802.4.

45
IEEE Standard 802
TOKEN RING MAC SUBLAYER PROTOCOL

• Frame Structure Components -
• Frame status
• A bit - the intended receiver saw the packet
• C bit - the receiver copied the packet into its
  buffers.
• Serves as acknowledgment.
• Priorities -
• Token gives priority of that token - a sender
  must wait for token of correct priority. The
  access control byte (of the token or data frame)
  has reservation bits. As frame goes by, a
  requester can say it wants the token at that
  priority the next time around.

46
IEEE Standard 802
TOKEN RING MAC SUBLAYER PROTOCOL

• RING MAINTENANCE
• Monitor station oversees the ring, but on failure
  any station can become monitor.
• CLAIM_TOKEN is a request to become the new
  monitor.
• Monitor oversees
• .Lost token management - If timer says token not
  seen in a while, produce new one.
• .Orphan frames - (Frame on ring, but sender
  crashes before draining frame.) Sets "monitor"
  bit in access control byte. If this bit seen as
  set the next time around, then something is
  wrong.
• .Garbled frame - Monitor drains the frame and
  issues new token.
• .Delay time - Ensures enough delay so whole token
  fits on ring.
• Broken rings handled by any station who thinks
  neighbors unreachable. Uses BEACON control type.
• Token management handled by the monitor so not
  de-centralized. Management easier, but
  susceptible to berserk behavior.

47
IEEE Standard 802
COMPARISONS OF 802.3, 802.4, AND 802.5

• In great scheme of things, differences are small.
  All three have approximately same technology and
  speed.

48
IEEE Standard 802
IEEE 802.2 Logical Link Control

• LLC
• For when a reliable error-controlled
  flow-controlled data link protocol is required.
• Also hides differences inherent in the 802.3/4/5
  from the network layer.
• Three possible options
• o Unreliable datagram service.
• o Acknowledged datagram service.
• o Reliable connection-oriented service.

Destination Address
Source Address
Control
Information
Set mode/Information/ Acknowledge/Poll
LLC Protocol Entity
LLC Protocol Entity
Set mode/Information/ Acknowledge/Poll
49
BRIDGES
Overview
This is one way that networks are connected
together. Bridges operate in the data link
layer, and so dont have the intelligence to do
much address resolution. What we will talk about
here - Translation from one LAN type to another.
Given a MAC address, how does a packet get to
its destination.

• 4.1 The Channel Allocation Problem
• 4.2 Multiple Access Protocols
• 4.3 IEEE Standards for LANs
• 4.4 Bridges
• 4.5 High Speed LANs

50
BRIDGES
The Big Picture

• Hub or repeater just electronic amplification.
• Bridges operate with active Data Link
  Layer. Can convert between different
  physical/data link types. Way to connect
  multiple LANs.
• Routers operate at Network layer -
  they read and depend on a specific protocol.
• Protocol Converters are able to convert from one
  network layer type to another.
• Detour on why to have multiple LANs -
• Organizations have different LANs (802.3/4/5) to
  meet various needs.
• Cost - may make the cabling less expensive.
• To carry a combined load heavier than any one LAN
  could do.
• Total distance more than 2.5 Km.
• Bridges can act as firewalls, to partition
  against errant hardware.
• LANs broadcast everything on the LAN to all
  stations. May want to prevent this from
  happening for some data. A bridge partitions off
  these messages.

51
BRIDGES
The Big Picture

• How they work -

52
BRIDGES
From 802.x to 802.y

• Issues -
• Each LAN type has its own frame format. Bridges
  take off one type and put on another. 4.36
• LANs don't necessarily run at the same speed, so
  must reject or buffer the data
• Two input LANs feeding one output LAN.
• Each LAN type has it's own maximum data length.
  So bridges must do framing in order to translate.
• Network layers may time out because they expect
  the destination to ACK within a given time all
  this translating slows down the transmission.
• All LAN types don't carry the same information
• o Priority bits.
• o Acknowledgment bits.
• In essence, the LAN standards are incompatible.

53
BRIDGES
Bridge Types

• In addition to translating packets, bridges also
  route packets between source and destination.
  Its this function we now turn to.
• Transparent Bridges and Source Routing Bridges
  are two competing and mutually exclusive ways of
  routing packets.

54
BRIDGES
Bridge Types

• TRANSPARENT BRIDGES
• Also called Spanning Tree Bridges -
• Goals
• "Perfect" transparency. No one needs to do
  anything. It just works.
• No hardware or software configuration required.
• No switches, no routing tables.
• Stateless (or as stateless as possible.)
• Promiscuous mode
• Accepts all packets from all LANs attached to
  the bridge.
• If destination is on incoming LAN, discard the
  packet.
• Otherwise, forward the packet.
• Use table (hashed) in bridge to determine choice
  of the LAN for forwarding.

55
BRIDGES
Bridge Types

• Parallel Redundant Bridges
• Here two or more bridges are used for
  reliability.
• Problems with infinite flooding.
• Solution is to overlay the Backward Learning
  policy with a virtual loop-free topology.
• A Spanning Tree Bridge does this.
• Note how paths are reduced.

56
BRIDGES
Bridge Types

• Parallel Redundant Bridges
• (cont)
• Algorithm is as follows
• Choose one bridge to be root of tree
• (lowest unique serial number wins.)
• Each bridge determines cost of the path from the
  root bridge to each of its ports. (The root path
  cost.) Cost determined by number of segments and
  the bit rate of those segments.
• Determine the root port - for a bridge, which of
  its ports has the lowest root path cost.
• Determine the designated bridge - the bridge that
  will handle requests for a particular LAN (even
  though that LAN may have several bridges attached
  to it.) Selection based on smallest path cost
  from the segment to the root bridge.
• Tree includes every LAN but not necessarily every
  bridge.
• Continues to check for topology changes.

57
BRIDGES
Bridge Types

• SOURCE ROUTING BRIDGES
• Used by IBM/rings. Here the sender holds ALL
  knowledge of how a packet should be routed.
• The sender of a frame
• Knows if destination is on its own LAN.
• Sets a bit alerting bridges destination NOT on
  its LAN.
• Places, in header, the exact path frame will
  follow.

58
BRIDGES
Bridge Types

• SOURCE ROUTING
• BRIDGES(cont)
• In the figure above, the Path from A to D is
  L1, B1, L2, B2, L3.
• The Bridge looks for bit set.
• Scans route - is the incoming LAN number followed
  by the number of the bridge doing the looking?
  If so, forward the frame, otherwise reject it.
• Can be done in software, hybrid, hardware.
• If the source doesn't know the route, it sends a
  "discovery frame" that goes to every LAN in the
  network. The destination replies and each bridge
  along the way puts its ID in that reply. The
  source then knows all that it needs. This
  discovery produces lots of excess packets.

59
BRIDGES
Bridge Types

• COMPARISON OF 802 BRIDGES

60
High Speed LANs
Overview
xxxx

• 4.1 The Channel Allocation Problem
• 4.2 Multiple Access Protocols
• 4.3 IEEE Standards for LANs
• 4.4 Bridges
• 4.5 High Speed LANs

61
SUMMARY
Here we want to know how to handle broadcast
networks. As compared to point to point
networks, a major issue is handling arbitration
when there is competition for the network. This
is the bottom sublayer of the Data Link Layer.
This Chapter is especially relevant for LANs. 4.1
The Channel Allocation Problem How to allocate
a single channel among multiple users. 4.2
Multiple Access Protocols How to handle
contention for the use of a channel. 4.3 IEEE
Standards for LANs How do the protocols of the
last sections apply to real systems. Here we
talk about the actual standards in use. 4.4
Bridges Ways of connecting networks
together. 4.5 High Speed LANs Directions in
high speed networks.

• 2.1 Theoretical Basis For Data Communication
• What every sophomore EE knows !!! How much data
  can be put on a wire? What are the limits
  imposed by a medium?
• 2.2 Transmission Media
• Wires and fibers.
• 2.3 Wireless Transmission
• Radio, microwave, infrared, unguided by a medium.
• 2.4 The Telephone System
• The system invented 100 years ago to carry
  voice.
• 2.5 Narrowband ISDN
• Mechanisms that can carry voice and data.