ATM NETWORKS

1
1. INTRODUCTION

2
NETWORK PARADIGMS
Bandwidth (Mbps)
ATM LANS Gigabit Ethernet
ATM
1000
Voice, Image, Video, Data
Fast Ethernet FDDI
100
SMDS (DQDB)
10
Ethernet/ Token Ring/ Token Bus
Frame Relay
1
X.25
Distance
LAN
MAN/WAN
3
NETWORKING EVOLUTION
Disjoint Networks
Traditionally

• Voice Telephone networks
• Data Computer networks and LAN
• Video Teleconference Private corporate
  networks
• TV Broadcast radio or cable networks

These networks are engineered for a specific
application and are ill-suited for other
applications, e.g., the traditional telephone
network is too noisy and inefficient (capacity
limited) for bursty data communication.
Data networks are not suitable for voice and
video traffic because they cannot satisfy
the time-sensitivity of these traffic types.
4
Old Fashioned
Broadband ISDN
WIDE AREA
WIDE AREA
X.25 Internet
B-ISDN A Single Unifying Technology
PSTN
LOCAL
LOCAL
PBX
LAN
Video, Image, Voice and Data
Voice
Data
Before Integration
After Integration
1.6-64 kbps gt Narrowband 1.5-45 Mbps gt
Wideband gt 45 Mbps gt Broadband
5
GOAL INTEGRATION
(One Network Carrying Multimedia Traffic)
Broadband Integrated Services Network (Data,
Voice, Video, Still Image)
Source
Destination

• WHY INTEGRATION?

Voice Traffic (based on ATT data, 1993) 125 to
130 Million Calls/day x 5 min/call x 64 kbps
28.8 Gbps 1 / 1000th of fiber capacity
6

• Updated statistics for 1998
• Average calls per business day 272.2 million
• Average calls per day 226.2 million
• Average length per business call 2.5 min.
• Average length per consumer call 8.0 min.

Suppose 200 Million x 24 hours/day x 64 kbps
12.8 Tbps Bottomline Gigantic Capacity of fiber
cannot be utilized only by Voice Traffic!!!

• FURTHER REASONS
• Convergence of computer and communications
  technologies
• Integration could offer efficiencies (lower
  cost) and
• support of new applicatons
• Single network management maintenance
• No duplication of cables, plants (since one
  physical network)
• gt Less costs

7
INTEGRATION PROBLEMS
Integration is not easy because different
applications have different performance
requirements.
Medium Speed Data
Telemetry Telecontrol Telealarm Voice Telefax
Hifi Sound
Video Telephony
Low Speed Data
High Speed Data
High Quality Video Video Library Video Education
Very High Speed Data
8
INTEGRATION PROBLEMS

• Voice
• Video
• Data
• Still
• Image

• 64 Kbps (Bandwidth Demand)
• 2 min. (End-to-End Delay on the Average)
• (Request-Transfer Cycle)

• gt 140 Mbps (BW Demand)
• 60 min. (E2E Delay on the Average)

• Extremely variable (BW Demand)
• Extremely variable (E2E Delay on the Average)
• e.g., long telnet sessions short finger
  requests

• 1-50 Mbps (BW Demand)
• 1 Sec. (E2E Delay on the Average)
• (Cannot be seen as data traffic because it may
  have real-time
• nature, e.g. medical image retrieval geographic
  databases)

9
ATM NETWORKS
Asynchronous Transfer Mode (ATM)
A new multiplexing and a new switching
technique to realize the Broadband Integrated
Services Digital Networks (B-ISDN)
Asynchronous Packet transmission is not
synchronized to a global (network) clock!!!
10
Multiplexing
Defines the means by which multiple streams of
information share a common physical
transmission medium.
Sources
Physical Link (Channel)
Mux.
N
Shares single output between many inputs.
Sinks (Destinations)
Demux.
Demux has one input which must be distributed
among outputs.
11
Multiplexing Techniques
Multiplexing Techniques
Frequency Division Multiplexing
Time Division Multiplexing
Synchronous TDM
Asynchronous TDM (Statistical Multiplexing)
(STM)
(ATM)
12
STM
STM Circuit Switching used for Telephone
Networks (also for N-ISDN) (Time
Division Multiplexing) (Classical TDM)
Information is transferred with a certain
repetition frequency. e.g., 8 bits every
125?sec for 64kbps 1000 bits every
125?sec for 8Mbps
From Nyquists Sampling Theorem 4kHz voice signal
requires 8000 samples/sec 8
bits/sample One 8 bit sample every 125?sec
64kbps (Golden Rule of Tel. Networks) (DS0
Channel)
Basic unit of repetition frequency is called a
TIME SLOT.
13
STM
Time Slot
. . .
. . .
. . .
n
n
1
1
2
2
Periodic Frame (one cycle)
Start of each Frame
L sec (? bits can be transmitted)

• Each slot is assigned to a particular call. The
  call is identified by the position of the slot.
  When the user is assigned a slot, it owns a
  circuit. The user uses the same slot within
  consecutive frames.
• If a user is not transmitting data in its own
  slot, that time slot remains reserved (nobody
  else can transmit there).

14
ATM
ATM In ATM, the MUX takes a packet (one or more
packets) appends a header (5 Bytes) and
transmits based on statistical characteristics of
the sources. MUX can take according number of
cells from a source transmit.
15
ATM
e.g., from a video source 10 cells can be taken,
while from voice source 1 cell because of their
bandwidth demands.
Each packet must have a header (control
information)!!
No SYNCHRONIZATION!! (No network clock!)
16
STM vs. ATM
Simple Example
17
STM vs. ATM
STM (Synchronous Transfer Mode)
ATM (Asynchronous Transfer Mode)
18
STM vs ATM
(STM)
(ATM)
MUX (Multiplexer)
19
STM

• Advantages
• No overhead in packetization.
• Constant repetition of frames low delay
  jitters.
• Easy to maintain synchronization between sender
  receiver.

• Disadvantages
• Limited flexibility
• BW (bandwidth) allocation at 64kbps modularity
• Inefficient for VBR (Variable Bit Rate) traffic
• Long connection set-up delays.
• Complex switching system.

20
ATM Advantages

• Flexible BW-Allocation (sources with widely
  different bit rates).
• Accommodate bursty sources (e.g., VBR)
• Asymmetrical link bandwidths

• A wide range multimedia traffic types.
• High efficiency due to statistical multiplexing.
• Allows Quality of Service (QoS) guarantees.
• Simple routing (small buffers).
• Simple switching.

21
ATM Disadvantages

• Overhead for cell header.
• 48 5 (bytes) 40Bits overhead ? 10 overhead
• Requests fast switching technology (need new
  switches).
• Complex scheduling algorithms needed.
• Connection set-up signaling overhead.
• Traffic management problem.
• Difficult to reroute virtual circuits.
• Jitter problem.

22
SWITCHING TECHNIQUE

• Takes multiple instances of a physical
  transmission medium containing multiplexed info
  streams and rearrange the info streams between
  input output.
• In other words, information from a particular
  physical link in a specific multiplex position is
  switched to another output physical link.

23
SWITCHING TECHNIQUE
24
Switch Architectures

• Single Bus
• Self-Routing (Blocking)
• Multiple Bus
• Self-Routing (Non-Blocking Queueing
• - Internal Queueing
• - Input Queueing
• - Output Queueing
• (Shared Buffer ?
• Theoretically optimal ?
• Achieves maximum throughput!)
• Actual ATM switches have combination of input,
  output, and internal queueing.
• The way how these functions are implemented,
  where in the switch these functions are located,
  will distinguish one switching solution from
  another.

25
SWITCHING
Question
Why could existing switch architectures (circuit-s
witches for voice, packet-switches for data
) NOT be used for ATM ?
Reasons
1. High speed at which the switch must operate
(155-622 Mbps now on Gigabit levels) 2.
Statistical behavior of ATM streams passing
through the switch 3. ATM has small fixed cell
size limited header functionality
26
ATM SWITCHING
Combines Space Time Switching Principles!!
27
ATM Basic Switching Principle
28
ATM Switching

• All cells which have header a or (c or d ) on
  incoming I1 are switched to O1 and their header
  is translated (switched) to value p or (k or q
  ).
• All cells with a header c (or b or e) on link Ij
  are also switched to outlet On, but their header
  gets values k (or y or r).
• Remark On each incoming outgoing link
  individually, the values of the header are
  unique, but identical headers can be found on
  different links, e.g., c on link I1 and Ij.
• Realization
• Routing info is contained in the header (label),
  not explicit address.
• Explicit addressing is not possible because of
  short fixed size cell.
• A physical Inlet/Outlet, characterized by a
  physical port number.
• A logical channel on the physical port
  characterized by a VCI and/or VPI.

Routing Tables Must be set up in
advance (signaling phase)
Either pre-defined or dynamically allocated
29
ATM SWITCHING
Routing Info
Basic Functions Space Switching
(Routing) Header Switching
Queueing
Why Queueing? Suppose 2 cells from different
Inlet (I1 In) arrive simultaneously at ATM
switch and are destined to the same outlet O1.
Thus, they cannot be put on the output or the
outlet at the same time ? buffering, i.e., to
store the cells which cannot be served. (No
pre-assigned time slots, statistical multiplexing)
Remark The way these 3 functions are
implemented, where in the switch these functions
are located, will distinguish one switching
solution from another.
30
ATM NETWORK
31
ATM CELL STRUCTURE
8 7 6 5 4 3 2 1
1 2 3 4 5 53
Octet

• Octets are sent in increasing order
• ? 1,2,3
• Within an octet the bits are sent
• in decreasing order ? 8,7,6,5,4 ...

HEADER (5 octets)
PAYLOAD (48 octets)
User Network Interface (UNI) Cell Structure
Network Network Interface (NNI) Cell Structure
8 7 6 5 4 3 2 1
8 7 6 5 4 3 2 1
1 2 3 4 5 53
1 2 3 4 5 53
GFC
VPI
VPI
VCI
VPI
VPI
VCI
GFC Generic Flow Control VPI Virtual Path
Identifier VCI Virtual Channel Identifier PT
Payload Type PR Priority HEC Header Error
Control
VCI
VCI
VCI
VCI
PT
PT
PR
PR
HEC
HEC
PAYLOAD (48 octets)
PAYLOAD (48 octets)
32
ATM Interfaces
33
ATM Network Interfaces (Detailed)
34
ATM Cell

• Generic Flow Control (GFC) (4 Bits) (only at
  UNI)
• It provides flow control information towards the
  network. It allows a multiplexer to control the
  rate of an ATM terminal. Currently, no standard.
  0s are used for this field.
• Routing Field (VPI/VCI)
• 24 Bits (8 Bits for VPI, 16 for VCI) at UNI.
• 28 Bits (12 for VPI, 16 for VCI) at NNI.
• VPI/VCI have only local significance only they
  identify the next destination.
• Remark
• Each physical UNI to support not more than 28256
  VPs. NNI ? 212 4096 VPs. Each VP can support
  216 65,636 VC on UNI and NNI.
• Payload Type Field (PT) (3 Bits)
• PT indicates whether the cell contains users
  data, signaling data or maintenance information.
• Cell Loss Priority (CLP) (1 Bit)
• CLP indicates the priority of the cell. Lower
  priority cells are discarded before higher
  priority cells when congestion occurs.
• Remark
• If CLP 1 ?Cell has low priority ? dropped in
  heavy load.
• If CLP 0 ?Cell has high priority ? not
  discarded.
• Header Error Control (HEC) (8 Bits)
• HEC detects and corrects errors in the header.
  (i.e., single Bit Error Correction or
  Multiple-Bit Error Detection). The info field is
  passed through the network intact, with no error
  checking or correction. ATM relies higher
  protocols for this purpose.

35
ATM CELL

• PAYLOAD TYPE (PT)
• First Bit ? 0 ? User
  Information
• First Bit ? 1 ? Network
  Management or Maintenance Function
• Second Bit ? Whether CONGESTION has been
  experienced or not.
• Third Bit ? known as AAU
  (ATM-User-to-ATM-User) used in AAL5 to convey
  information between end users.
• Contents EFCI (Explicit Forward
  Congestion Indication)
• 0 0 0 ? User Data Cell Congestion No
  (EFCI0) AAU0
• 0 0 1 ? User Data Cell Congestion No
  (EFCI0) AAU1
• 0 1 0 ? User Data Cell Congestion Yes
  (EFCI1) AAU0
• 0 1 1 ? User Data Cell Congestion Yes
  (EFCI1) AAU1
• 1 0 0 ? Segment Operation and Maintenance
  (OAM) (F5) Cell
• 1 0 1 ? End-to-End (OAM) Flow F5 Cell
• 1 1 0 ? Resource Management Cell
• 1 1 1 ? Reserved for Future Function

36
Pre-Assigned (Pre-Defined) (Reserved) Header
Values
37
Cell Types

• Idle Cell Inserted and extracted by PHY in
  order to adapt the cell flow rate at the
  boundary between ATM layer PHY layer to the
  available payload capacity of the transmission
  system.
• Valid Cell has a header with no error or which
  has been corrected by the
• HEC verification process.
• Invalid Cell has a header that has errors that
  have not been modified by the HEC verification
  process (discarded at PHY layer).
• Assigned Cell provides service to an
  application using ATM layer service.
• Unassigned Cell Not an assigned cell. Does not
  contain any useful
• information.

38
Cell Types (Cont.)
Source
Destination
Assigned Cell
Assigned Cell
Upper Layers
Upper Layers
Unassigned Cell
Unassigned Cell
ATM Layer
ATM Layer
SAP
SAP
PHY Layer
PHY Layer
Idle Cell
Valid Cell
Invalid Cell
Idle Cell
Network
Trash
39
Cell Types (Cont.)
Difference Idle Cells vs. Unassigned
Cells Unassigned Cells ? Visible to ATM
PHY layer. Idle Cells ? Visible only to PHY
layer ? not to ATM layer. Unassigned Cells are
sent whenever there is no information available
at the sender. It allows full asynchronous
operation of sender/receiver. Idle Cells are
inserted by the PHY layer in order to match the
transmission rate to the transmission system or
for other PHY layer purposes. Octet
1 Octet 2 Octet 3 Octet 4 Octet 5
0 . . . 0 0 . . . 0 0 . . . 0 0 . .
. 0 0 . . . 0 Each octet of Info. field of
an idle cell is filled with 01101010.
40
The Size of the ATM Cell(WHY 48 5 53 BYTES?)

1. Transmission Efficiency
2. Delays
3. Implementation Complexity

41
1) Transmission Efficiency
where L is the information size of the packet in
bytes and is the header size of
the packet in bytes.
The longer the info. field, the higher is the
efficiency for the same header size. A header of
4 or 5 Bytes is typical value for ATM
cell. (Assumption All packets are completely
filled.)
x
42
2. DELAYS

• Packetization Delay (Segmentation)
• Transmission Delay (depends on the distance
  between both endpoints). (Range 4-5?sec per km
  depends on the transmission medium).
• Switching Delay
• Queueing (Buffering) Delay
• Depacketization Delay (Reassembly)
• Queues are necessary to avoid massive loss of
  cells. Delay varies with the load of the network
  and is determined by the behavior of queues.
• Conclusions
• The queueing delays increase with the size of
  information field.
• The end-to-end delay must be below 24 ms to
  avoid ECHO problems for voice traffic!!!

43
EXAMPLE Packetization Delay (Segmentation)
Transmission of 64 kbps voice traffic over
ATM. Voice signal is sampled 8000 times per
second, which gives rise to 8000 bytes/sec or 1
byte every 125 usec. If the packet size is 16
bytes, then it will take (16125) usec or 2ms to
fill up a packet. If the packet size is 64
bytes, then it will take (64125) usec or 8ms to
fill up a packet. So the smaller the packet size,
the less the delay to fill up a packet. The
packetization delay could be kept small if a
packet is partially filled however, this will
lead to under-utilization of the network capacity.
44
EXAMPLE Time for Header Conversion
The longer the packet, the more time the switch
has to do the header conversion. Consider an ATM
switch with OC-3 capacity, i.e., 155 Mbps. If the
cell size is 53 bytes, then a maximum of about
365 566 cells can arrive per second. This
translates to 2.7 usec per cell, i.e., assuming
that cells arrive back-to-back, a new cell
arrives approximately every 2.7 usec. This means
that the switch has 2.7 usec available to carry
out the header conversion. Suppose a cell size of
10 bytes. A maximum of about 1 937 500 cells can
arrive per second, i.e., if cells arrive
back-to-back, a new cell arrives approximately
every 0.5 usec. The switch has then only 0.5 usec
for the header conversion.
45
3) Implementation Complexity

• Two parameters play a role in determining the
  complexity of a system
• The speed Transmission (Processing) Time (P)
• (Cell Size/Data Rate)
• The number of required bits (MEMORY M))
• The number of cells (BUFFER SIZE IN CELLS)
  multiplied by the (CELL SIZE).
• Tradeoff ? Memory Size and Processing Speed
• To guarantee a certain limit on the cell loss
  ratio, a number of cells must be provided per
  queue. This number is independent of the cell
  size. So the larger the cell size, the larger the
  queue in bits will be (e.g., doubling the cell
  size will also double the memory requirements).

46
IMPLEMENTATION COMPLEXITY
On the other hand, for every cell, the header
must be processed. This processing must be
performed in one cell time, so the longer the
cell size, the larger the available time and the
lower the speed requirements of the system. In
Figure we show the speed and memory size in
function of the cell size, if the system operates
at 150 Mbps and if the queue is dimensioned for
50 cells (the header is 4 Bytes).
47
Explanation of the FIGURE Assume 50 Cells Cell
Size 16 Header 4 20 Bytes (Low)
Cell Size 256 Header 4
260 Bytes (High) Memory (M)
Cell Size Buffer Size in Cells 2050 1000
Bytes 8000 Bits (Low)

2605013000 Bytes104000 Bits (High) Processing
Speed (P) Transmission Time Cell
Size/Data Rate160 Bits/150106 bps 1
musec (L)

2080/150106 bps 13.8 musec (H)
48

1. We see that for a cell of 16 bytes, we need only
   about 8000 bits for the memory, but the header
   processing of each cell must be performed in less
   than 1 ?sec.
2. For a cell of 256 bytes, we need already more
   than 64,000 bits for a single queue. But we have
   about 15 ?sec for the header processing of a
   single cell.
3. However, as seen in Figure, the speed is not the
   most critical issue, since in 1 ?sec (in case of
   16 Bytes) HIGH processing can be achieved so the
   limiting factor is the memory space requirement.

49
FINALLY THE RESULT Contradicting factors are
contributing to the choice of the cell size.
However, a value between 32 and 64 bytes is
preferable. Europe was in favor of 32 bytes
(because of the requirement for echo cancellers
for voice) where US and Japan were in favor of
64 bytes because of higher transmission
efficiency. Finally gt a compromise of 48 Bytes
reached at CCITT meeting in June 1989.
50
Variable vs. Fixed Length Packets

• Facts to consider in decision
• Transmission Bandwidth Efficiency
• Achievable Switching Performance
• (i.e., the switching speed vs. complexity)
• The Delay

51
Transmission Bandwidth Efficiency

• Fixed Packet Length,
• where Number of useful information in
  bytes
• Information Size of the
  Packet in bytes
• Header size of the packet in
  bytes
• represents the smallest
  integer larger than or equal to

52

• This efficiency is optimal for all information
  units which are multiples of the packet
  information size, i.e.,
• Optimal Case (Substitute the above value into the
  prev. one)

53

• 96 144 192
  240 288 336
  Xnumber of useful
• 
  
  information bytes

• has a sawtooth shape (Opt. L48H5)

54

• The efficiency depends very much on the useful
  information bytes to be transmitted
• If the number of useful information bytes is
  large, the optimal achievable efficiency is
  approached
• Only if the number of useful information bytes
  is small, this efficiency is rather low.
• So, the distribution of the number of useful
  information bytes to be transmitted largely
  determines the efficiency.

55
Different Applications

• Voice
• Since voice is a CBR (Constant Bit Rate) service,
  we can take the option at the sending terminal
  only to transmit a packet when it is completely
  filled (therefore, introducing a packetization
  delay).
• So, the efficiency can reach the optimal
  achievable value, if packets are completely
  filled which then puts limitation on the packet
  size in order to limit the packetization delay.

56
Different Applications

• Video
• Where fixed bit rate video coding techniques are
  used, this service can be considered as a CBR
  service, again reaching the optimal efficiency
• Where variable bit rate video coding techniques
  are used, it may occasionally happen that packets
  are not completely filled.
• However, a typical video image contains thousands
  of bytes, so the optimal achievable efficiency
  will be very closely approached.

57
Different Applications

• Data
• Distinguish low speed and high speed data.
• Low speed applications, e.g., keyboard input
  small information units must be considered, so
  the efficiency is rather small (around 10)
• High speed applications, e.g., file transfer,
  image transfer for CAD, etc the very long
  information field (e.g., file, image, etc) can be
  sent into fixed packets giving rise to an
  efficiency very close to the optimal efficiency,
• e.g., for 1000 bytes, the efficiency is 89,
  instead of an 90.5 in the figure)

58

• Remark
• Since traffic in a broadband network will largely
  be composed of video, high speed data, and voice,
  the overall transmission efficiency approaches
  the optimal, even if fixed length packets are
  used.

59

• Variable Length Packets

• Here the overhead is determined by the
  header and the flags to delimit the packets,
  e.g., 6 bits in HDLC, plus in addition, some
  stuffing bits to ensure proper flag recognition.
• Also, add to the header, a length
  indicator, determining the length of the packet
• where is the specific packet header
  overhead mentioned above

60

• In Figure, we assume 5 bytes and 2 bytes of
  . We see the transmission efficiency can be
  very high (close to 100) for very long packets
• Remark For practical reasons such as buffer
  dimensioning, delay, the max variable length
  packets must be limited to a certain threshold.

61
CONCLUSIONS

• The transmission efficiency of variable length
  packets is better than that of fixed length
  packets
• However, in broadband networks, this gain of
  transmission efficiency is rather limited since
  the main traffic constituting broadband services
  will consist of a combination of voice, video,
  bulk data transfer.

62
SWITCHING SPEED AND COMPLEXITY
Two factors for complexity of ATM switch
implementation

• Speed of Operation
• Queue Memory Size Requirements

A) Speed of Operation
Header processing
Let us assume that header functions are the same
for fixed and variable length case. For fixed
length packets, available time to perform all
functions is fixed, (e.g., 2.8 ?sec in the 48 5
bytes solution at 150 Mbps.)
For variable length packets, the available time
depends on the worst case (i.e., the smallest
packet), so the speed requirements are much
higher (e.g., to perform same functions for a 5
5 byte packet at 150 Mbps only 553 ns are
available.).
63
SWITCHING SPEED AND COMPLEXITY
B) (Queue) Memory Management
Fixed Length Packets ? Memory management
system can assign memory stocks with the same
size, namely, the same size of the packets. This
operation is simple and management of free memory
list is easy.
Variable Length Packets ? Memory management
system must be able to assign memory stocks in
multiples of bytes so that algorithms like find
best fit, find first fit,, can be used.
Memory management is complex. CONCLUSION
Regarding speed of operation and queue memory
size Fixed Length Packet Size is
preferred!!!! In 1988, fixed packet size has been
accepted for ATM.
64
FACTS on ATM TECHNOLOGY

• Provides a way of linking a wide range of devices
  (from telephones to computers) using the seamless
  network
• Also removes the distinctions between LAN, MAN
  and WAN)
• Combines packet and circuit switching
• It can be sent on any physical media (copper,
  fiber). Wide range of transmission speed.
• Scalable
• Allows QoS parameters (voice, video, still image,
  etc.)
• Supports any type of traffic
• Allows sources of different bit rates
• Uses fixed size packets called CELLS.

65
FACTS on ATM TECHNOLOGY

• No error protection or flow control on hop base
• Header functionality is reduced
• Information field is very small.
• Operates in Connection-Oriented Mode
• Supports Connectionless Mode

66
HISTORY of ATM
67
ATM FORUM TECHNICAL COMMITTEES

• Traffic Management
• Signaling
• Physical Layer
• Testing
• B-ICI
• LAN Emulation
• SAA (Service Aspects Applications) (VTOA)
• Network Management
• P-NNI
• Multiprotocol over ATM (MPOA)
• Residential Broadband

68
ATM NETWORKS