Asynchronous Transfer Mode (ATM)

• A Connection-oriented network providing Quality of Service guarantees

– developed in the mid-1980’s to combine packet switching with TDM
– had been hoped it would be the end-user to end-user network of choice
» to provide broadband services beyond ISDN
– TCP/IP Internet became the de facto standard instead
– ATM now used as part of the network infrastructure
» e.g. in enterprise networks, IP over ATM, Voice over ATM etc.
– claimed benefits :
» high performance hardware switching
» dynamic bandwidth allocation for bursty traffic
» class-of-service support for multimedia applications
» common LAN/WAN architecture via LAN emulation
» scalability in speed and network size
1
• An ATM network is a collection of endpoints, switches and interfaces :
Private Private
UNI ATM
network
X X
U NI
c
Private bli Public ATM
NNI X Pu network A

X X
Public
UNI
X NNI
B-ICI Public ATM
network B
– UNI : user-network interface X
Public
» connects end systems to switches UNI
X X
– NNI : network-network interface
» connects two switches
– B-ICI : broadband intercarrier interface
» connects two public switches from different service providers
– provides Permanent Virtual Connections (PVCs) e.g. leased lines,
and Switched Virtual Connections (SVCs) on demand
2
• ATM Technology
– all information formatted into fixed-length cells of 48 bytes + 5 bytes header
» ensures time-critical information not adversely affected by long packets
– header organised for efficient hardware switching
» carries payload type information, virtual-circuit identifiers, CRC etc.
– information from separate flows converted into cells and then multiplexed
» cells queued and transmitted according to some scheduling strategy
» no time-slot reservation as in TDM
- cells flow asynchronously

Voice
Data
packets MUX
Wasted bandwidth
Images

TDM
4 3 2 1 4 3 2 1 4 3 2 1

ATM `
4 3 1 3 2 2 1

3
– ATM is connection-oriented
» a connection setup required prior to transfer of cells
» identifies a path through the network that can provide the required service
- a virtual channel connection (VCC)
» the VCC has a chain of local identifiers each used at a switch (VCIs)
- each input port to a switch uses its own private set of VCIs
- used to index into its routing table to find the next hop output port

1 voice 67
Switch 1
…

video 67 2
5 video 25 voice 32 25 N 75
32 1 67 data 39 3
32 3 39

…
data 32 video 61
6 61 2 67
…

video 75 N
N

» high speed switching relatively straightforward, 155Mbps streams typical

- fixed sized cell with standard headers
4
– Aggregation into virtual paths:
» virtual channel connections (VCC) sharing a common path through the network

VP3 a
VP5
a
b ATM ATM
c ATM ATM b
Sw DCC Sw Sw c
d 1 2 3
e

VP2

VP1
ATM d
Sw e
4

» VCCs a, b, c aggregated into VP3 after switch 1

- and to VP5 after the Digital Cross Connect to switch 2
» VCCs d, e aggregated into VP2 after switch 1
- and to VP1 after the DCC to switch 4
» VCCs b, c could have been aggregated after switch 2 to switch 3
» the DCC here switches aggregated virtual paths, not individual VCCs
5
– Two levels of identifiers
» Virtual Channel Identifier (VCI) and Virtual Path Identifier (VPI)
Virtual Paths
Physical Link

Virtual Channels

» all switches along a virtual path switch on the basis of the VPI alone
» VCIs only used at the end of a virtual path
– the VCI/VPI structure supports scalability to very large networks
– DCCs allow network path topology reconfiguration under software control
» similar to SONET ring reconfigurability
– network managers can allocate bandwidth to virtual paths
» and to any degree of granularity

6
– ATM Cells:
UNI cell NNI cell
GFC VPI VPI
VPI
VPI
VCI VCI
PT CLP PT CLP
HEC HEC

Payload Payload
(48 bytes) (48 bytes)

» GFC (Generic Flow Control) : local functions e.g. identifying multiple stations at
end-points
- not used in practice
» no GFC in NNI header allows larger trunks between public ATM switches
» VPI : 8 or 12 bits; VCI : 16 bits
» PT (Payload Type) : payload contains user data/control data
» CLP (Cell Loss Priority) : lower priority cells get discarded first when congested
» HEC (Header Error Control) : CRC-8 (x8+x2+x+1) over first 4 bytes of header
- detection and/or single bit correction
7
• Reference Model:
Management plane

Plane management
Control plane User plane

Layer management
Higher Higher
Application
layers layers

Presentation
ATM adaptation layer
Session

Transport

Network ATM layer

Data Link

Physical Physical layer

» user plane deals with data transfer, flow control, error recovery etc.
» control plane deals with signalling to set up, manage and release connections
» management plane deals with network resources and coordination of other
planes

8
 ATM layer
– Physical layer:
» transmission convergence sublayer Transmission
convergence
- ATM cell boundaries tracked Physical layer
sublayer

- checking of header checksums Physical medium

dependent
- insertion and removal of idle cells sublayer

» physical medium dependent layer

Physical medium
- cells are converted into a bitstream
- cells packaged into frames for the medium e.g. SONET, FDDI, STP etc.
– ATM layer:
» concerned with sequenced transfer of cells in connections across the network
» accepts 48-byte blocks from the ATM Adaptation layer and adds 5-byte header
User information User information

AAL AAL

ATM ATM ATM ATM

PHY PHY PHY PHY

…

End system Network End system

9
– ATM Adaptation layer :
» responsible for providing different applications with the appropriate support
- several AAL types are defined – AAL1 to AAL5
» converts higher level SDUs into 48-byte blocks for ATM layer

Voice A/D AAL

s1 , s2 … cells
Digital voice samples

Video A/D … Compression AAL

cells
compressed
picture frames
frames

Data AAL
Bursty variable-length cells
packets

» provides support for the protocol layer directly above

- e.g. to provide a reliable stream service, if needed (not if TCP was the layer above)
- cell sequence numbering
- segmentation and reassembly support
10
• Service Classes
– Constant Bit Rate (CBR)
» rate constant for the whole period of the connection
» for traffic sensitive to delay
- e.g. voice, video, TV, circuit switching emulation
– Variable Bit Rate, non-real-time (VBR-NRT)
» for bursty sources with no rigorous timing requirements
- rate varies with time depending on rate at which application produces data
- e.g. multimedia e-mail, transaction processing
– Variable Bit Rate, real-time (VBR-RT)
» similar to VBR-NRT for applications sensitive to cell-delay variation
- e.g. voice with speech activity detection
– Available Bit Rate (ABR)
» allows sources to make use of any bandwidth that is still available
- e.g. file transfer, e-mail
– Unspecified Bit Rate (UBR)
» for everything else, including TCP/IP – lowest tariff
11
• Traffic Descriptors
– parameters which the source must specify when negotiating a connection
– traffic will be policed and enforced to conform with the parameters
– Peak Cell Rate (PCR)
» rate in cells/second that the source is never allowed to exceed
» inverse of minimum inter-cell arrival time
– Sustainable Cell Rate (SCR)
» average cell rate in cells/second produced by the source over a long period
– Maximum Burst Size (MBS)
» number of consecutive cells that may be transmitted by the source at the PCR
» used as the bucket size parameter for the enforcement algorithm
– Minimum Cell Rate (MCR)
» minimum cell rate in cells/second that the source is always allowed to send
– Cell Delay Variation Tolerance (CDVT)
» tolerable level of cell delay variation in a given connection

12
• Negotiable Quality of Service parameters
– Cell Loss Ratio (CLR)
» ratio of number of lost cells to total transmitted cells
- lost in the network due to congestion and/or buffer overflow
» specified as an order of magnitude in the range 10-1 to 10-15
» extent to which CLR can be negotiated depends on buffer allocation
strategies available in the network

– Cell Transfer Delay (CTD)

» delay experienced by a cell between network entry and exit
» includes propagation delays, queuing delays at intermediate switches and
processing times at queuing points
» since cells each experience different delays, CTD specified by a probability
density function
» maximum CTD can be negotiated by which some fraction (1-α) of the cells
will be delivered
- where α is some appropriately small value

13
 probability density
α

D0 Peak-to-Peak CDV

Dmax

– Cell Delay Variation (CDV)

» the total delay encountered by cells in a connection
- excluding fixed delay D0
» peak-to-peak delay can currently be negotiated
» network switches have only limited control of the variance of CTD values
- so range of negotiable CDV value is also limited

14
– relevant parameters for service classes:

Class of Service CBR VBR-NRT VBR-RT ABR UBR

CLR 9 9 9 9 U
CTD 9 U 9 U U
CDV 9 9 9 U U
PCR 9 9 9 U 9
SCR U 9 9 U U
MBS U 9 9 U U
flow control U U U 9 U

– Connection Admission Control (CAC)

» ensures that new virtual connections are assigned to links with sufficient
bandwidth to meet the committed levels of Quality of Service
- algorithms not standardised – left up to equipment suppliers
– Policing
» a Generic Cell Rate Algorithm (GCRA) defined
- equivalent to the leaky bucket scheme
15
• Connection establishment
– setup, call proceeding, connect and connect acknowledgment messages:

Source Transit Destination

Source A Destination B
Switch Switch Switch
setup
setup
setup
call proceeding setup
call proceeding
call proceeding call proceeding

connect
connect
connect
connect connect ack
connect ack
connect ack connect ack

release
release
release complete release
release complete release
release complete
release complete

• ATM addressing
– has its own system : telephony-oriented or private end-system addresses
16