Data-link Layer (Part 2)

1
 Reading assignment
 Chapter 5.5-5.11
 Due Wednesday, Oct. 8
 Homework #2
 Due Monday, Oct. 15

2
 Last class
 Data-link layer overview
 Data-link layer functions
 This class
 Specific data-link layers and devices

3
 Specific data-link layers
 Ethernet (802.3)
 Token Ring (802.5)
 WiFi (802.11)
 PPP
 ATM
 X.25
 Frame relay
 Specific data-link layer devices
 Hubs
 Bridges
 Switches

4
“dominant” LAN technology:
 First practical local area network, built at Xerox PARC in 70’s
 Cheap: $20 for 100Mbs!
 Simpler, cheaper than token LANs and ATM
 Kept up with speed race: 10, 100, 1000 Mbps

Metcalfe’s Ethernet
sketch

5
 Carrier sense
 Check to see if active transmission
 Collision detect
 Sender checks for collision; wait and retry
 Adaptive randomized wait to avoid collisions

6
Sending adapter encapsulates IP datagram (or other network layer protocol
packet) in Ethernet frame

 Preamble – 8 bytes
 7 bytes with pattern 10101010 followed by one byte with pattern 10101011
 Used to synchronize receiver, sender clock rates
 Src/Dst Address – 6 bytes
 Globally unique, allocated to manufacturers
 All adapters listening receive frame, discard if not destined for itself
 Type – 2 bytes
 Indicates higher layer (network) protocol being used, mostly IP these days
 http://www.cavebear.com/CaveBear/Ethernet/type.html
 Data – 46 to 1500 bytes
 CRC – 4 bytes
 Checked at receiver, dropped if doesn’t match
 CRC-32 (x32+x26+x23+x22+x16+x12+x11+x10+x8+x7+x5+x4+x2+x+1)

7
if packet
then {
A: sense channel
if idle
then {
transmit and monitor the channel;
if detect another transmission
then {
abort and send jam signal;
update # collisions;
delay as required by exponential backoff algorithm;
goto A
}
else {done with the frame; set collisions to zero}
}
else {wait until ongoing transmission is over and goto A}
}

8
 Packet?
No

Sense Send Detect

Carrier Collision

Yes
Discard
Packet
b=CalcBackoff();
attempts < 16 wait(b);
attempts++;
attempts == 16

9
 If deterministic delay after collision, collision
will occur again in lockstep
 If random delay with fixed mean
 Few senders  needless waiting
 Too many senders  too many collisions
 Exponentially increasing random delay
 Infer senders from # of collisions
 More senders  increase wait time

10
Exponential Backoff:
 Goal: adapt retransmission attempts to

estimated current load

 heavy load: random wait will be longer
 first collision: choose K from {0,1}; delay is K x
512 bit transmission times
 after second collision: choose K from {0,1,2,3}…
 after ten or more collisions, choose K from
{0,1,2,3,4,…,1023}

11
 What if two people
sent really small
packets
 How do you find
collision?
 Must have a
minimum packet
size

12
 Min packet length > 2x max prop delay
 If A, B are at opposite sides of link, and B starts one
link prop delay after A
 Jam signal
 Jam network for 32-48 bits after collision, then stop
sending
 Ensures that everyone notices collision

13
 Propagation delay
 1Km, c in cable = 60% * c in vacuum = 1.8 x
10^8 m/s
 1000/1.8 x 10^8 ~= 5 x 10^-6 = 5us
 5us * 3Mbps = 15bits in flight!
 Modern 10Mb Ethernet {
 2.5km=maximum network segment length, 10Mbps
 ~= 12.5us delay
 +introduced repeaters (max 5 segments)
 worst case – 51.2us round trip time!

14
 Slot time = 51.2us = 512bits in flight
 After this amount, sender is guaranteed sole access to link
 51.2us = slot time for backoff
 What about scaling? 100Mbit, 1Gbit...
 Make network smaller?
 Solution for 100BaseT
 Make min pkt size larger?
 512bits @ 1Gbps = 512ns
 512ns * 1.8 * 10^8 = 92meters
 Gigabit ethernet uses collision extension for small pkts

15
 Ethernet unstable at high loads
 Peak utilization = 1/e = 37%
 Peak throughput worse with
 More hosts – more collisions needed to identify
single sender
 Smaller packet sizes – more frequent arbitration
 Longer links – collisions take longer to observe, more
wasted bandwidth

16
 10: 10Mbps; 2: under 200 meters max cable length
 thin coaxial cable in a bus topology

 repeaters used to connect up to multiple segments

 repeater repeats bits it hears on one interface to its other interfaces: physical layer device only!

17
 10/100 Mbps rate; latter called “fast ethernet”
 T stands for Twisted Pair cabling
 Nodes connected to hubs or switches in a “star
topology”
 Max distance from node to Hub is 100 meters
 Smart hubs
 Disconnect “jabbering adapter
 Managed hubs (monitoring, statistics)

18
 Use standard Ethernet frame format
 Allows for point-to-point links and shared
broadcast channels
 In shared mode, CSMA/CD is used; short
distances between nodes to be efficient
 Uses hubs, called here “Buffered Distributors”
 Full-Duplex at 1 Gbps for point-to-point links

19
 Packets broadcast around ring
 Token “right to send” rotates around ring
 Fair, real-time bandwidth allocation
 Every host holds token for limited time
 Higher latency when only one sender
 Higher bandwidth
 Point to point links electrically simpler than bus

20
 4 Mbps
 max token holding time: 10 ms (limits frame
length)

• SD, ED mark start, end of packet

• AC: access control byte:
– token bit: value 0 means token can be seized, value 1 means data
follows FC
– priority bits: priority of packet
– reservation bits: station can write these bits to prevent stations with
lower priority packet from seizing token after token becomes free

21
• FC: frame control used for monitoring and maintenance
• source, destination address: 48 bit physical address, as in
Ethernet
• data: packet from network layer
• checksum: CRC
• FS: frame status: set by dest., read by sender
– set to indicate destination up, frame copied OK from ring
– DLC-level ACKing

22
 Failure modes
 Token rings – network unusable
 Ethernet – node detached
 Good performance in common case
 Volume  lower cost  higher volume ….
 Adaptable
 To higher bandwidths (vs. FDDI)
 To switching (vs. ATM)
 Completely distributed, easy to maintain/administer
 Easy incremental deployment
 Cheap cabling, etc

23
 Wireless LANs: untethered (often mobile)
networking
 IEEE 802.11 standard:
 MAC protocol
 unlicensed frequency spectrum: 900Mhz, 2.4Ghz

• Basic Service Set (BSS) (a.k.a.

“cell”) contains:
– wireless hosts
– access point (AP): base
station
• BSS’s combined to form
distribution system (DS)
24
 Ad hoc network: IEEE 802.11 stations can
dynamically form network without AP
 Applications:
 “laptop” meeting in conference room, car
 interconnection of “personal” devices
 battlefield
 IETF MANET
(Mobile Ad hoc Networks)
working group

25
802.11 CSMA sender
- if sense channel idle for DIFS sec.
then transmit entire frame (no collision
detection)
-if sense channel busy
then backoff (random, exponential)
802.11 CSMA receiver
if received OK
return ACK after SIFS
802.11 CSMA others
 NAV: Network Allocation
Vector
 802.11 frame has transmission time field
 others (hearing data) defer access for
NAV time units

26
 Same as previous mode
but with explicit channel
reservation
 sender: send short RTS:
request to send
 receiver: reply with
short CTS: clear to send
 CTS reserves channel for
sender, notifying
(possibly hidden) stations
 avoid hidden station
collisions
27
 RTS and CTS short:
 collisions less likely, of
shorter duration
 end result similar to
collision detection
 IEEE 802.11 alows:
 CSMA
 CSMA/CA: reservations
 Polling from AP

28
 one sender, one receiver, one link: easier than
broadcast link:
 no Media Access Control
 no need for explicit MAC addressing
 e.g., dialup link, ISDN line
 popular point-to-point DLC protocols:
 PPP (point-to-point protocol)
 HDLC: High level data link control (Data link
used to be considered “high layer” in protocol
stack!)

29
 http://www.rfc-editor.org/rfc/rfc1548.txt
 packet framing: encapsulation of network-layer datagram
in data link frame
 carry network layer data of any network layer protocol
(not just IP) at same time
 ability to demultiplex upwards
 bit transparency: must carry any bit pattern in the data field
 error detection: (no correction)
 connection liveness: detect, signal link failure to network
layer
 network layer address negotiation: endpoint can
learn/configure each other’s network address
30
 no error correction/recovery
 no flow control

 out of order delivery OK

 no need to support multipoint links (e.g.,

polling)
Error recovery, flow control, data re-ordering
all relegated to higher layers!|

31
 Flag: delimiter (framing)
 Address: does nothing (only one option)
 Control: does nothing; in the future possible
multiple control fields
 Protocol: upper layer protocol to which frame
delivered (eg, PPP-LCP, IP, IPCP, etc)
 info: upper layer data being carried
 check: cyclic redundancy check for error
detection

32
 “data transparency” requirement: data field
must be allowed to include flag pattern
<01111110>
 Q: is received <01111110> data or flag?

 Sender: adds (“stuffs”) extra < 01111110> byte

after each < 01111110> data byte
 Receiver:
 two 01111110 bytes in a row: discard first
byte, continue data reception
 single 01111110: flag byte

33
flag byte
pattern
in data
to send

flag byte pattern plus

stuffed byte in
transmitted data

34
Before exchanging network-
layer data, data link peers
must
 configure PPP link (max.

frame length,
authentication)
 learn/configure network

layer information
 for IP: carry IP Control
Protocol (IPCP) msgs
(protocol field: 8021) to
configure/learn IP
address

35
 Encrypt data at a layer below network layer
 IPsec only works for IP packets
 PP2P, L2F, L2TP
 Allow encryption at data-link layer
 Works for all network layer packets of all types
using PPP
 Often tunneled using IP packets
 IP(Tunnel) | PP2P/L2F/L2TP | Original Network
Packet

36
 1980s/1990’s standard for high-speed (155Mbps to 622 Mbps and
higher) Broadband Integrated Service Digital Network architecture
 Take strengths of IP, learn from its shortcomings
 Packet switching good
 Packet switching without explicit network-level connections and
reservations bad
 Design new network to address emerging applications while allowing
for efficient support for non-real-time data applications
 Goal: integrated, end-end transport of carry voice, video, data
 meeting timing/QoS requirements of voice, video (versus
Internet best-effort model)
 “next generation” telephony: technical roots in telephone world
 packet-switching (fixed length packets, called “cells”) using
virtual circuits

37
 adaptation layer: only at edge of ATM network
 data segmentation/reassembly
 roughly analagous to Internet transport layer
 ATM layer: “network” layer
 cell switching, routing
 physical layer
38
 ATM Adaptation Layer (AAL): “adapts” upper
layers (IP or native ATM applications) to ATM
layer below
 AAL present only in end systems, not in
switches
 AAL layer segment (header/trailer fields, data)
fragmented across multiple ATM cells
 analogy: TCP segment in many IP packets

39
Different versions of AAL layers, depending on
ATM service class:
 AAL1: for CBR (Constant Bit Rate) services, e.g. circuit
emulation
 AAL2: for VBR (Variable Bit Rate) services, e.g., MPEG
video
 AAL5:
User data for data (eg, IP datagrams)

AAL PDU

ATM cell

40
 AAL5: low overhead AAL used to carry IP
datagrams
 4 byte cyclic redundancy check
 PAD ensures payload multiple of 48bytes
 large AAL5 data unit to be fragmented into 48-byte
ATM cells

41
 Service: transport cells across ATM network
 analagous to IP network layer

 very different services than IP network layer

Guarantees ?
Network Service Congestion
Architecture Model Bandwidth Loss Order Timing feedback

Internet best effort none no no no no (inferred

via loss)
ATM CBR constant yes yes yes no
rate congestion
ATM VBR guaranteed yes yes yes no
rate congestion
ATM ABR guaranteed no yes no yes
minimum
ATM UBR none no yes no no

42
 VC transport: cells carried on VC from source to
dest
 call setup, teardown for each call before data can flow
 each packet carries VC identifier (not destination ID)
 every switch on source-dest path maintain “state” for each
passing connection
 link,switch resources (bandwidth, buffers) may be allocated
to VC: to get circuit-like perf.
 Permanent VCs (PVCs)
 long lasting connections
 typically: “permanent” route between to IP routers
 Switched VCs (SVC):
 dynamically set up on per-call basis

43
 Advantages of ATM VC approach:
 QoS performance guarantee for connection mapped
to VC (bandwidth, delay, delay jitter)
 Drawbacks of ATM VC approach:
 Inefficient support of datagram traffic
 one PVC between each source/dest pair does not
scale (N*2 connections needed)
 SVC introduces call setup latency, processing
overhead for short lived connections

44
 5-byte ATM cell header
 48-byte payload
 Why?: small payload -> short cell-creation delay for
digitized voice
 halfway between 32 and 64 (compromise!)

Cell header

Cell format

45
 VCI: virtual channel ID
 will change from link to link thru net
 PT: Payload type (e.g. RM cell versus data cell)
 CLP: Cell Loss Priority bit
 CLP = 1 implies low priority cell, can be
discarded if congestion
 HEC: Header Error Checksum
 cyclic redundancy check

46
Vision: end-to-end
transport: “ATM from
desktop to desktop”
 ATM is both a network
technology and a data-
link layer technology
Reality: used to connect IP
backbone routers
 “IP over ATM”
 ATM as switched link
layer, connecting IP
routers

47
 Classic IP only
 3 “networks” (e.g., LAN segments) IP over ATM
 MAC (802.3) and IP addresses
• replace “network” (e.g.,
LAN segment) with ATM
network
• ATM addresses, IP
addresses
ATM
network

Ethernet Ethernet
LANs LANs
48
Issues:
• IP datagrams into ATM
network
ATM AAL5
PDUs
• from IP addresses
to ATM addresses
– just like IP Ethernet
LANs
addresses to 802.3
MAC addresses!

49
 at ATM network source edge:
 IP layer finds mapping between IP, ATM dest address (using
ARP)
 passes datagram to AAL5
 AAL5 encapsulates data, segments to cells, passes to ATM
layer
 ATM network: moves cell along VC to destination
 at ATM network destination edge:
 AAL5 reassembles cells into original datagram
 if CRC OK, datgram is passed to IP

50
 Start with ATM switch
 Rip out ATM signaling
 Rip out ATM routing
 Add IP routing software
 Add Flow classifier to map IP flow to underlying ATM virtual circuit
 Operation
 Upon arrival of first packet in flow
 Assign IP flow to a virtual circuit ID (label) using IP routing software
 Label locally significant
 Subsequent flows
 Switched in hardware after flow classified
 IP packet forwarding done as label index lookup O(1) versus IP route lookup
O(log n)
 http://pnewman.org/papers/infocom96.pdf
 Subsequent versions add label or flow ID distribution to downstream nodes
 Akin to local routing algorithm

51
 Later generalized as MPLS (multi-protocol
label switching)
 “Layer 2 ½”
 Not tied to ATM
 Extensible to IPv6
 http://www.rfc-editor.org/rfc/rfc3031.txt
 Used as a tool for traffic engineering
 http://www.rfc-editor.org/rfc/rfc2702.txt

52
Like ATM:
 wide area network technologies
 virtual circuit oriented
 origins in telephony world
 can be used to carry IP datagrams
 can thus be viewed as Link Layers by IP
protocol

53
 X.25 builds VC between source and destination for each
user connection
 Per-hop control along path
 error control (with retransmissions) on each hop using
LAP-B
 variant of the HDLC protocol
 developed when bit error rates over long-haul copper
links were orders of magnitude higher
 per-hop flow control using credits
 congestion arising at intermediate node propagates to
previous node on path
 back to source via back pressure

54
 X.25: reliable in-sequence end-end delivery
from end-to-end
 “intelligence in the network”
 built for dumb terminals accessing mainframes
 IP: unreliable, out-of-sequence end-end
delivery
 “intelligence in the endpoints”
 gigabit routers: limited processing possible
 2000: IP wins

55
 Designed in late ‘80s, widely deployed in the
‘90s
 Frame relay service:
 no error control
 no flow control
 End-to-end congestion control

56
 Designed to interconnect corporate customer
LANs
 typically permanent VC’s: “pipe” carrying
aggregate traffic between two routers
 switched VC’s: as in ATM
 corporate customer leases FR service from
public Frame Relay network (eg, Sprint, ATT)

57
 flags address data CRC flags

 Flag bits, 01111110, delimit frame

 address:
 10 bit VC ID field
 3 congestion control bits
 FECN: forward explicit congestion notification
(frame experienced congestion on path)
 BECN: congestion on reverse path
 DE: discard eligibility
 Precursor to IP DiffServ and ECN

58
 Committed Information Rate (CIR)
 defined, “guaranteed” for each VC
 negotiated at VC set up time
 customer pays based on CIR
 DE bit: Discard Eligibility bit
 Edge FR switch measures traffic rate for each VC; marks
DE bit
 DE = 0: high priority, rate compliant frame; deliver at “all
costs”
 DE = 1: low priority, eligible for discard when congestion
 Precursor to IP DiffServ
 Can be used to support higher layer QoS mechanisms

59
Q: Why not just one big LAN?
 Limited amount of supportable traffic: on
single LAN, all stations must share bandwidth
 limited length: 802.3 specifies maximum cable
length
 large “collision domain” (can collide with
many stations)
 limited number of stations: 802.5 have token
passing delays at each station

60
 Effectively a physical layer device
 Multi-port repeater
 Repeater operating at bit level
 Repeat received bits on one interface to all other interfaces
 Hubs can be arranged in a hierarchy (or multi-tier
design), with backbone hub at its top

61
 Each connected LAN referred to as LAN segment
 Hubs do not isolate collision domains: node may
collide with any node residing at any segment in LAN
 Hub Advantages:
 simple, inexpensive device
 Multi-tier provides graceful degradation: portions of
the LAN continue to operate if one hub malfunctions
 extends maximum distance between node pairs
(100m per Hub)

62
 single collision domain results in no increase in
max throughput
 multi-tier throughput same as single segment
throughput
 individual LAN restrictions pose limits on
number of nodes in same collision domain and
on total allowed geographical coverage
 cannot connect different Ethernet types (e.g.,
10BaseT and 100baseT)

63
 Link Layer devices: operate on Ethernet
frames, examining frame header and
selectively forwarding frame based on its
destination
 Bridge isolates collision domains since it
buffers frames
 When frame is to be forwarded on segment,
bridge uses CSMA/CD to access segment and
transmit

64
 Bridge advantages:
 Isolates collision domains resulting in higher total
max throughput, and does not limit the number of
nodes nor geographical coverage

 Can connect different type Ethernet since it is a store

and forward device

 Transparent: no need for any change to hosts LAN

adapters

65
 Bridges filter packets
 same-LAN -segment frames not forwarded onto
other LAN segments
 Forwarding:
 how to know which LAN segment on which to
forward frame?
 looks like a routing problem (more shortly!)

66
67
 Not recommended for two reasons:
- single point of failure at Computer Science hub
- all traffic between EE and SE must path over CS segment

68
 Basic idea: build cache of which nodes are
downstream of which ports
 Monitor source address on all packets bridge
forwards
 Insert reverse path to source into cache
 What to do with unknown sources?
 Flood network

69
 bridges learn which hosts can be reached through which
interfaces: maintain filtering tables
 when frame received, bridge “learns” location of
sender: incoming LAN segment
 records sender location in filtering table
 filtering table entry:
 (Node LAN Address, Bridge Interface, Time Stamp)
 stale entries in Filtering Table dropped (TTL can be 60
minutes)

70
 filtering procedure:
if destination is on LAN on which frame was received
then drop the frame
else { lookup filtering table
if entry found for destination
then forward the frame on interface indicated;
else flood; /* forward on all but the interface on
which the frame arrived*/
}

71
Suppose C sends frame to D and D replies back with
frame to C

• C sends frame, bridge has no info about D, so floods to

both LANs
– bridge notes that C is on port 1
– frame ignored on upper LAN
– frame received by D
72
• D generates reply to C, sends
– bridge sees frame from D
– bridge notes that D is on interface 2
– bridge knows C on interface 1, so selectively forwards
frame out via interface 1

73
 for increased reliability, desirable to have redundant,
alternate paths from source to dest
 with multiple simultaneous paths, cycles result - bridges
may multiply and forward frame forever
 solution: organize bridges in a spanning tree by disabling
subset of interfaces

Disabled

74
 Switches
 “multi-port bridge”
 Each port acts as a bridge
 Each port determines MAC addresses connected to
itself
 Master list within switch determines forwarding
behavior

LAN 1 LAN 2

75
 Higher link bandwidth
 Point to point electrically simpler than bus
 Much greater aggregate bandwidth
 Separate segments can send at once
 Data backplane of switches typically large to support
simultaneous transfers amongst ports
 Challenge
 Learning which packets to copy across links
 Avoiding forwarding loops

76
 layer 2 (frame) forwarding,
filtering using LAN
addresses
 Switching: A-to-B and A’-to-
B’ simultaneously, no
collisions
 large number of interfaces
versus bridges (which
typically have only two)
 often: individual hosts, star-
connected into switch
 Ethernet, but no collisions!

77
 cut-through switching: frame forwarded from
input to output port without awaiting for
assembly of entire frame
 slight reduction in latency
 combinations of shared/dedicated,
10/100/1000 Mbps interfaces

78
 Dedicated

Shared

79
 both store-and-forward devices
 routers: network layer devices (examine network layer headers)
 bridges are Link Layer devices
 routers maintain routing tables, implement routing algorithms
 bridges maintain filtering tables, implement filtering, learning
and spanning tree algorithms

80
Bridges + and -
+ Bridge operation is simpler requiring less processing
bandwidth
- Topologies are restricted with bridges: a spanning tree
must be built to avoid cycles
- Bridges do not offer protection from broadcast storms
(endless broadcasting by a host will be forwarded by a
bridge)

81
Routers + and -
+ arbitrary topologies can be supported, cycling is
limited by TTL counters (and good routing
protocols)
+ provide firewall protection against broadcast storms
- require IP address configuration (not plug and play)
- require higher processing bandwidth

 bridges do well in small (few hundred hosts) while

routers used in large networks (thousands of hosts)

82
 principles behind data link layer services:
 error detection, correction
 sharing a broadcast channel: multiple access
 link layer addressing, ARP
 various link layer technologies
 802.3 Ethernet
 802.5 Token-ring
 802.11 LANs
 PPP
 ATM
 X.25, Frame Relay
 hubs, bridges, switches

83