Local Area Networks (LAN)

Local Area Network (LAN)

 Can be seen as a broadcast domain
 The set of machines such that a broadcast message from
any machine will reach all other machines
 Machines are on the “same network”
 That does not mean machines can only broadcast in
LANs, a machine can send a message for another specific
machine (unicast) or a group of machines (multicast)
 Characteristics
 Usually (but not necessarily) spread over a small
geographical area Unicast = one to one
multicast = one to many
 Relatively high data rate compared to WAN broadcast = one to all

 Single administrative management

 Topologies
 Many specific physical topologies earlier: Bus, star, ring
 Switched LANs nowadays for arbitrary physical topologies
 Specifications at physical and data link layer mostly
 We have already seen most of the general issues there
 We will study Ethernet, the de-facto standard protocol for
LANs now
 History of Ethernet
 Developed in early 70’s in Xerox PARC by Metcalfe and Boggs
 DIX standard – Digital/Intel/Xerox standardized 10 Mbps
Ethernet in early 80’s (final one called Ethernet-2)
 IEEE 802 body – standardized various physical layer/MAC
combinations, and a single Logical Link Control (LLC) on top of
them (1985)
 IEEE 802.2 : LLC
 IEEE 802.3 : CSMA/CD
 IEEE 802.4 : Token ring
 Many others since then
Ethernet-2 does not do any error control
or flow control. It does frameing, deframing
and error detection, medium access (mac)

 Specifications at data link and physical layers

 Ethernet-2 specifies only MAC sublayer and physical
layer
 IEEE 802.2 + one of IEEE 802.3/802.4 etc. specifies
complete data link and physical layer
 We will see what this means shortly
IEEE 802.3 specifically with CSMA/CD

 Working group of IEEE dealing with LANs

 A collection of standards defining various Ethernet types and
related issues
 Multiple types depending on speed, media type etc.
 10Base-5 : 10Mbps, thickwire coaxial cable
 10Base-2 : 10Mbps, thinwire coax or cheapernet
 10Base-T : 10Mbps, UTP
 10Base-F : 10Mbps, optical fiber
 100Base-TX : 100Mbps, twisted pair
 100Base-FX : 100 Mbps, optical fiber
 1000Base-LX : 1 Gbps, optical fiber
 1000Base-T : 1 Gbps, UTP
 10GBASE-T : 10 Gbps, UTP
 10GBASE-SR : 10 Gbps, optical fiber
 There are many other types not covered here
  All of them share the same basic frame format, difference is
mostly in MAC and physical layer details like parameter values,
encodings, media, connector etc.
 802.3 primarily specify CSMA/CD for half duplex operations
over shared media CSMA/CD was ment for shared media
(half duplex operation)

 But modern day LANs are all switched LANs with full duplex
operation, CSMA/CD is not needed mostly (why?)
Switched LANs also use full duplex operation, which means that devices can transmit and receive data simultaneously on separate
channels. This eliminates the need for CSMA/CD (Carrier Sense Multiple Access with Collision Detection), which was a technology
used in older Ethernet LANs to manage access to the shared communication medium and avoid collisions between data packets.

CSMA/CD was designed for half-duplex Ethernet LANs, where devices could not transmit and receive simultaneously on the same
communication channel. In such LANs, devices had to listen for activity on the network before transmitting their own data to avoid
collisions with other devices. If a collision occurred, CSMA/CD detected it and triggered a backoff algorithm to retry transmission.

With full duplex operation, devices on a switched LAN can transmit and receive data simultaneously without worrying about
collisions, because each device has its own dedicated communication channel. This greatly improves the efficiency and
performance of the LAN, making it more suitable for modern data-intensive applications.
Relation Between the Standards
 Standards now are from IEEE
 “Ethernet LAN” and “IEEE 802.3 LAN” used interchangeably
 Frame format for both standards the same except for small
difference
 TCP/IP implementations use original Ethernet-2 frame format,
no LLC (network layer directly uses Ethernet frames)
 Both types can coexist on the same LAN (will discuss how later)
Minimum Frame Size
 A frame must take more than 2τ time to send (τ = max. one-way
propagation delay) to detect collision
 An example scenario:
 Two machines A and B located at the far ends of the cable
 A starts sending a frame at time 0
 Frame almost at B at time τ –δ
 B starts sending a frame at time τ – δ
 Collision occurs at time τ
 Jam signal gets back to A at time 2τ
 If A has finished transmitting the frame (before 2τ time), the collision is missed
 Minimum frame size of 10 Mbps Ethernet = 64 bytes
 Slot time = time to transmit the minimum sized frame = 512bits/10Mbps =
51.2 microseconds for 10 Mbps Ethernet
 If a station transmits for 1 slot time and does not detect any collision, it will be
able to finish its transmission
 What about Fast Ethernet (100 Mbps) and Gigabit Ethernet (1 Gbps)?
 What if it is <46, then pad it with 0's

Ethernet Frame Format max so that nobudy takes the channel

for too long
who this data is for

7 1 6 6 2 46-1500 4
IEEE
PA SFD DA SA LEN LLC PDU PAD FCS 802.3

8 6 6 2 46-1500 4

PA DA SA Type DATA PAD FCS Ethernet

PA: Preamble --- 10101010s for synchronization

SFD: Start of frame delimiter --- 10101011 to start frame
DA: Destination MAC address
SA: Source MAC address
LEN: Length --- number of data bytes
Type: Identify the higher-level protocol
LLC PDU + Pad: minimum 46 bytes, maximum 1500
FCS: Frame Check Sequence --- CRC-32

receiver has to know which softwware

on sender side has sent. Which higher layer
has given the data. ex lets say IP
 All machines, in physical layer will get 10 Mbps:
the message. 64 bytes (min frame size, same for back-
First FCS is done at each machine if there ward compatibility)
is error then frame is dropped else the 51.2 microseconds slot time (to pump
DA is checked if matches then ok else out those bits)
dropped
100Mbps:
# 64 bytes (min frame size, same for back-
How will receiver decide if it is padded ward compatibility)
or original data ? (>46 bit data then ok, 5.12 microseconds slot time (to pump
else if =46 then it may have padded bits) out those bits)

total propagation delay has to be smaller

as transmission faster so ack should
also be received faster.

#
How receiver will detect end of frame
(as data size is variable) ?

so reduce this distance

ideal state
minimal interframe gap,
this can be used to calculate
end of frame
#
How will the sender know the MAC
address of receiver ?
(dont worry about this now , there is
another protocol for this)
#
instead of particular MAC address if all
48 bits are 1111.... then this means for
everybody in LAN
MAC Address
24 bits 24 bits

Vendor Code Serial Number

0000.0c12.3456

MAC address is associated with NIC

NIC (network interface card)

ff.ff.ff.ff.ff.ff : Broadcast address

There are also multicast addresses
 MAC Address (contd.)
#
all mac address are unique

 The MAC sublayer defines a hardware address which is unique for

each LAN interface (NIC)
 The address is a 48-bit address, expressed as 12 hex digits
 Vendor code given to NIC manufacturers by IEEE
 Serial number given by vendor from its pool
 Hierarchical address ensures uniqueness
12 hex digits, also known as a 48-bit hexadecimal number, is a sequence of 12 characters that can be any of the following:

0, 1, 2, 3, 4, 5, 6, 7, 8, 9, A, B, C, D, E, F.

Hexadecimal numbers are often used in computing and digital systems to represent values, such as memory addresses or color
codes. Each hex digit represents four binary digits, or bits, which means that a 12 hex digit number represents 48 bits of information.
Interoperation Between Ethernet and
802.3
 All protocols type values standardized have values
greater than 1536
 Max. length field value in 802.3 frame = 1500
 If length/type field value ≤ 1500, it is a 802.3 frame,
else it is a Ethernet-2 frame
 It is also possible to carry protocols using type field
inside a 802.3 field (SNAP headers)
 Not discussed in this course
Some Terminologies
without repeaters you can put atmost
 Segment 100m distance = segment length

 Part of medium without any repeater

 One or more devices can connect to a segment
 Segments can be connected using repeaters
 Collision domain
 Set of devices such that simultaneous transmission by any
two devices will cause a collision
 One or more segments
IEEE 802.3i: 10Base-T
 10 Mbps, Baseband, Unshielded Twisted Pair (two pairs), Cat 3 or
better
 Max. segment length = 100m
 No. of machines/segment = 1
 Manchester encoding
 1-persistent CSMA/CD for transmission
 10 Mbps Ethernet is mostly not used for connecting computers
now
 But modified 10 Mbps Ethernet standards are being used in other
network-enabled applications for connecting devices
IEEE 802.3u: 100Base-TX
 Fast Ethernet (a group of 100 Mbps standards)
 100 Mbps, Baseband, Unshielded Twisted Pair, Cat 5 or better, or
Type 1 STP
 Max. segment length = 100 m
 4B/5B+MLT-3 encoding
 Other variations of Fast Ethernet also uses NRZI
 1-persistent CSMA/CD for transmission
IEEE 802.3 1 Gbps: 1000BaseLX
 1 Gbps, Baseband, on single or multimode optical fiber
 Specified in IEEE802.3z
 Two strands of fiber – one for transmit, one for receive
 Max distance varies between 550 m and 5 Km depending
on different factors
 8B/10B Encoding
 CSMA/CD (with some changes) for half duplex
Physical Interconnection
 Bus topology with multidrop lines
 One long connection with taps to connect machines
 One machine’s transmission can reach all machines
 The destination machine accepts based on address in frame
 Collision domain– all machines
 Inflexible wiring, hard to extend
 Think of a wiring already done and now you have to extend
your network to one more room in the building
 Star/Tree topology with hubs
 Hub: a central device with ports (4-port, 8-port, …)
 One machine can be connected to each port
 A frame received on one port is always broadcast to all
ports (irrespective of the destination)
 So logically still the same as a bus topology
 Collision domain = all machines
 Easy wiring, can extend easily by interconnecting hubs in
tree topology
 A hub just repeats the signal
 Operates at physical layer
 # layer 2 switches means switches
based on layer 2 information
# hub is a layer 1 device

layer 2 = data link layer

 Star/Tree topology with switches (Layer-2 switch)

 Switch: similar to hub with ports
 Difference: A frame received on one port is only sent on a
port to which the destination is connected (how?)
 So logically a point-to-point topology
 Collision domain = 1 machine
 Can extend easily by interconnecting switches in tree
topology same as hubs
 A switch also repeats the signal but looks at the MAC
address
 Operates at data link layer
 Almost all LANs are switched LANs now
Switched Ethernet
 Frames no longer broadcast always, sent to only port to
which destination is connected
 Separates the single collision domain of hub-based Ethernet
to multiple collision domains
 Allows more than one pair to communicate simultaneously
 Increases bandwidth available to each machine
 How does the switch know which machine is connected to
which port?
 Use a forwarding table
 Entries of the form <port, destination MAC addresses>
 Looks up destination MAC address and forwards to appropriate
port
 How is the table built?
 Switch learns!
 Suppose m/c A sends frame to m/c B
 Switch knows nothing initially, so broadcasts to all ports
 But switch now knows which port m/c A is connected to!
 If a frame comes for m/c A, it will be forwarded to only
A’s port
 Internal table built up as more and more machines
communicate, completely built up when every m/c has
sent at least one frame
 What if we connect more switches?
 Ex:You need to connect 16 machines (M1-M16), but you have
only 8-port switches
 Use 3 no. 8-port switches, S1, S2, S3
 S1 connects 7 machines (M1-M7), S2 connects 7 machines (M8-
M14)
 One remaining port of S1 connected to one port of S3, same for S2
 S3 connects 2 machines (M15 at Port 1, M16 at Port 2), plus have
one connection each from S1 and S2 (say at Port 3 and Port 4)
 S3’s table will eventually map M15 to Port 1, M16 to Port 2, M1-
M7 to Port 3, and M8-M14 to Port 4
 A frame from say M15 to M6 will be forwarded by S3 to Port 3
 S1 will then forward it to the port M6 is connected to (in local
table of S1)
 Learns the same way as for a single switch, just that entries in
table are of the form <port, <list of destination MAC
addresses>> what about, Bottle neck problem ?
 # switch :
if 1 sends to 4 then 2 can also send to 4
(for full duplex 1-4 and 4-1)
# in hub if 1 is sending to 4 nobudy can
send anything like 2 cant send to 3 , 3
cant send to 4.

# switch also repeats.

# when switch broad casts messege to all
ports it has learnt MAC address of all
addresses.
Spanning Tree Protocol
 For practical reasons, the interconnection between multiple
switches may not be a tree
 One switch’s failure can cut off a large number of machines
from each other
 So switches may be connected in an arbitrary physical
topology
 So there can be more than one path between two switches
 Why can this be a problem?
 A frame broadcast by a machine may go on circulating forever
 Solution:
 Allow arbitrary physical topology of switches, but create a
logical tree of switches out of it at any one time
 Spanning Tree Protocol and its variations (IEEE802.1d/1w)
Broadcasting can go on for ever = borad-
cast storm
 Spanning Tree Protocol (STP)
 Switches broadcast their switch ids to elect one switch as the
root switch (root bridge)
 By default the switch with the lowest MAC address
 Can be configured by the administrator also (why needed?)
 Each switch then finds the shortest “cost” path to the root
bridge
 The selected links form a rooted spanning tree
 Only the switch ports on the links in the tree are used for
forwarding packets (the port is in “forwarding” state)
 All other ports are put in “blocked” state, not used for
forwarding
 Control packets (Bridge Protocol Data Units (BPDU))
exchanged on all ports periodically to monitor the topology
 Reconfigure the links if a path is broken
Types of Switches
 Ethernet switches are commonly called L2 switches
 since they switch based on L2 address (MAC address)
 Unmanaged switch
 Can not be monitored or configured remotely
 Low cost
 Managed switch
 Can be monitored and configured remotely using network
management protocols
 Higher cost, but easier to maintain
 10/100 Mbps switches are still available, but most switches
from common OEMs are now gigabit switches
 Each port is of speed 1 Gbps (or higher for backend switches)
 Downward compatible usually to 10/100 Mbps cards
Bridges
 Connects more than one LAN segment
 LANs can be of same type (Transparent Bridging)
 No modification to content or format of frame
 No encapsulation
 Exact bitwise copy of frame
 MAC frames relayed, so does not need LLC
 May have buffering to meet peak demand
 LANs can be of different types (Non-Transparent Bridging)
 Frames may have different formats
 Needs to transform frames before sending to other LAN
 We consider Transparent Bridging only
Bridge Routing
 How does a bridge know which frame should go to which
LAN?
 Has a routing table similar to switches
 Entries of the form <port, <list of destination MAC addresses>>
 Looks up destination MAC address and forwards to appropriate
port
 How is the routing table built?
 Fixed and manually built – ok for small LANs
 Automatic: same concepts as address learning and STP for switches
 A transparent bridge looks very similar to a switch! Are they
the same? difference between switches and bridges:
switches: from anyone to all
A network bridge connects two local area
networks. A network switch, on the other
hand, connects multiple clients to a
network.
Auto-negotiation
 Allows two network devices to negotiate speed and other
parameters
 A 1000 Mbps network card connected to a 100 Mbps port
of a switch
 A half-duplex capable card connected to a full-duplex
capable switch
 Can be turned on or off
 Protocol allows devices to exchange their abilities, the
highest common abilities of the two devices is chosen
usually
 Mostly relevant for network devices connected with
copper (UTP/STP) wires.
LLC (Logical Link Control)
 IEEE 802.2
 Provides a common interface to higher layers for data
transmission
 Provides optional link control functionalities (error
control, flow control etc.) over MAC layers
 Common protocol over all IEEE 802.x MAC layers
We dont do this in ethernet since even
if link is made perfect there can be drops
at router etc. so anyways we need to do
at higher level, so why to do at lower
level.

MAC Frame Format for 802.3

PA SFD DA SA LEN LLC PDU PAD FCS

LLC
DSAP SSAP CTRL Data
PDU

which protocol should which protocol sent this

process this
LLC Header Fields
 SSAP - 8-bit address to identify Source Service Access Point (protocol)
above data link layer whose data is carried by the LLC PDU
 DSAP – 8-bit address to identify Destination Service Access Point
(protocol) above data link layer to whom the data in LLC PDU
will be given
 Control – 1 or 2 bytes
 Data – Actual data, variable sized (but multiples of bytes)
Example
 Modes of Operation
 Unacknowledged connectionless (Type 1)
 no acknowledgment, flow or error control; error
detection and discard at MAC level
 Connection-oriented (Type 2)
 Supports explicit connection
establishment/reset/termination, flow control,
sequencing, error control
 Acknowledged connectionless (Type 3)
 No connection establishment, but acknowledgements are
sent by receiver and retransmissions done by sender
 Frame formats to support Go-back-N or Selective Reject
ARQ
 7 bit sequence no. (frames numbered modulo 128)
 Exact frame formats skipped here
 IEEE 802.2 not used for IP networks
 Can use using something called SNAP header (will not
discuss), but not done
 IP networks work on Ethernet-II directly