M.

Vuskovic

ATM Networking

CS596

Chapter 1 ATM BASICS

Table of contents:

1.1 OVERVIEW 1.1.1 Why ATM? 1.1.2 Transmission Media 1.1.3 ATM Cells 1.1.4 Statistical TDM 1.1.5 Why the Cell Size is 53 Octets? 1.1.6 ATM Protocol Stack 1.1.7 Virtual Circuits 1.1.8 Guaranteed Quality of Service 1.1.9 LAN Emulation 1.1.10 Circuit Emulation 1.1.11 ATM Interfaces 1.2 PHYSICAL AND ATM LAYERS 1.2.1 Header Structure 1.2.2 Virtual Paths and Virtual Channels 1.2.3 Permanent Virtual Circuits 1.2.4 ATM Switching 1.2.5 Payload Type 1.2.6 ATM Protocol Stack 1.2.7 Cell Loss Priority 1.2.8 Header Error Control 1.2.9 Cell Delineation 1.2.10 Cell Scrambling 1.2.11 Repetitive Checking 1.3 ATM ADAPTATION LAYERS 1.3.1 Service Categories 1.3.2 Traffic Classes 1.3.3 ATM Adaptation Layer 1.3.4 ATM Adaptation Layer 1.3.5 ATM Adaptation Layer 1.3.6 ATM Adaptation Layer

1 2 3/4 5
Copyright C 1998, 1999 by Marko Vuskovic

1-2

OVERVIEW
Why ATM? To support any type of traffic: - burtsy data (to multimegabit rates: files, images, multimedia) - intermittent data (interactive systems, low rate, delay intolerant) - voice (sustained data rate, 64 kbps) - video (sustained data rate, multimegabit rates) To support transactions that use data, voice, and video simultaneously To provide high bandwidth, which can't be found in other technologies To provide a uniform architecture for fast LANs and scalable WANs of unrestricted sizes To provide bandwidth on demand (pay for use) To support multicast operations (video conferencing) To provide guaranteed quality of service To provide a unified approach in network management

LAN

PBX Router

Terminal/ service adapter

ATM
1.5 Mbps - 9 Gbps

Copyright C 1999 by Marko Vuskovic

1-3

OVERVIEW (Cont.)

ATM can be cost-effective for both, small and large networks, ranging from LANs to WANs, which can operate at T1/E1 to over 6 Gbps

Speed OC-129

OC-12

SONET/SDH
OC-3

ATM

45 Mbps

SMDS
1.5 Mbps

ISDN, X.25
64 kbps

FR

10

20

50

100

200

500

1000

Network size (number of sites)

[Used from: D. Spohn: "Data Network Design," McGraw-Hill, 1997]

SONET - SynchronousOptical Network SDH - Synchronous Digital Hierarchy ATM - Asynchronous Transfer Mode SMDS - Switched Multi-Megabit Data Services FR - Frame Relay X.25 - Protocols for packet-switched public data networks

Copyright C 1999 by Marko Vuskovic

1-4

OVERVIEW (Cont.)

Low speed packet switching networks can take bursty transactions at low throughput
Burstiness [peak/average] 1000

500
200 100 50 20 10 5 2 1 0.01 0.1 1 10 100 Circuit switching

SMDS X.25 FR

ATM

1000

Throughput [Mbps]

Circuit-switched networks are not very good for bursty traffic

[Used from: D. Spohn: "Data Network Design," McGraw-Hill, 1997]

Copyright C 1999 by Marko Vuskovic

1-5

OVERVIEW (Cont.)

Slow packet switching networks have large range of nodal delay

Range of nodal delay [msec] 100

SMDS is designed to support data only services, not well suited for voice

X.25

FR ATM SMDS

10

0.1

0.01

0.001 0.01

Circuit switching 0.1 1 10 100 1000

Throughput [Mbps]

Range of nodal delay for circuit-switched networks is very small - they have essentially constant nodal delay
[Used from: D. Spohn: "Data Network Design," McGraw-Hill, 1997]

Copyright C 1999 by Marko Vuskovic

1-6

OVERVIEW (Cont.)

Transmission Media The connections between DTEs (ATM interfaces) and ATM switches, as well as the connections between ATM switches are possible through a variety of transmission media.

ATM Switch

ATM Switch

DS (T1, T3, E1, E3) SONET/SDH (OC-3, OC-12, ...) STP (shielded twisted pair) UTP (unshielded twisted pair, category 5) Wireless Satelite

Copyright C 1999 by Marko Vuskovic

1-7

OVERVIEW (Cont.)
ATM Cells One of the main characteristics of ATM are small fixed-size cells (53 octets, 5 octets overhead, 48 octets payload). This gives the following advantages: predictable delay of cells can implement cell switching entirely in hardware smaller packetization delay, better support for voice and video 5 octets Overhead 48 octets Payload

Packetization Delay Its a consequence of not sending the packet before it is filled with data. 8 bit samples 125 sec

1 2 3 4 16 msec Packetizing 128 samples 1024 bit packet Delay of the first sample Packet has arrived (Now you can hear what was said 16+x ms ago) Propagation + processing delay (= x) 1 2 3 4

8 bit samples 1 2 3 6 msec Packetizing 48 samples 384 cell payload Delay of the first sample 1 2 3 4

Cell has arrived (delay only 6+x ms)

Copyright C 1999 by Marko Vuskovic

1-8

OVERVIEW (Cont.)

Statistical TDM Second important characteristic of ATM is statistical time division multiplexing (see Appendix B), also called: asynchronous TDM (which gives rise to the term "ATM").

Multiplexing is facilitated due to small fixed-size cells, resulting in smaller delays caused by sharing the transmission media among several sources.

Source A

A1

A2

Source B

B1 B1 A1 B1

B2

B2

Packet switching

A2

B2

Cell switching A1

B1

B2

A2

Copyright C 1999 by Marko Vuskovic

1-9

OVERVIEW (Cont.)

Why the Cell Size is 53 Octets? The cell size is determined as a trade off between packetizing delay and cell overhead. As shown later, the minimum required overhead (cell header) is 5 octets. The cell overhead (waste, "cell tax") is smaller at larger payload sizes. However, larger payload sizes produce larger packetization delays. This can cause echo in voice communication, and jerky motion in video. In addition the delay effect, a smaller cell size is preferred because of cell loss. There is larger probability to get bit errors in larger cells than in smaller cells. Losing a smaller cell due to error will not be noticed in voice communication. For example a 53-octet cell can contain 48 voice samples, which is only 48 x125 s = 6 ms of voice. The loss of such cell would be almost unnoticed. However a loss of cell with 32 ms of voice would be very disruptive. It is shown that the maximal payload size, at which there is no noticeable packetizing delay (echo and jerky motion) at 64 kpbs transmission, is 32 octets. This size however produces an overhead of 13.5 %. On the other hand a 64-octet payload would produce a small echo in voice communication, but the overhead would be only 7.8 %. USA and Japan proposed 64 octet payloads, while the rest of the world proposed 32 octet payloads. (USA and Japan argued that the small echo can be cancelled electronically at the receiver side.) In order to agree upon a unique cell size throughout the world, a compromise had to be reached:

Europe USA & Japan

64 + 32 = 48 2

Copyright C 1999 by Marko Vuskovic

1-10

OVERVIEW (Cont.)
ATM Protocol Stack The ATM protocol stack has three layers: AAL (ATM Adaptation Layer), ATM layer, and the physical layer. This stack doesn't fit exactly into the OSI reference model. For example, the ATM and AAL layers can't be interpreted as link and network layers respectively. What comes above the three ATM layers depends on the context of the ATM implementation (such as direct ATM, LAN emulation, IP over ATM, multiple protocol over ATM)

Adapts any traffic into the ATM format of 48-octet payload (currently, there are no applications that produce ATM cells directly.)

Checks and ensures some level of integrity with the packets/frames passed to AAL from higher layer protocols (link, network, transport,...); adds CS overhead. Also performs reverse operations on packets frames coming from the ATM layer. Service Specific CS (SSCS) Common Part CS (CPCS) Slices packets/frames into segments that can accommodate ATM cells; adds SAR overhead. Also does reverse operation: assembles cells coming from ATM layer into packets/frames

Convergence Sublayer (CS)

AAL
(ATM Adaptation Layer)

Segmentation and Reasembly (SAR)

ATM Physical
Adds cell header and passes cells to the physical layer. Perform cell switching and modification of cell headers at ATM switches.

Transmission Convergence

(TC)
Physical Media Dependent

Error control, cell delineation Physical frame generation. Mapping of cells into/from DS1, DS3, SONET/SDH frames.

(PMD)

Line encoding (AMI and B8ZS) and transmitting/receiving the bit stream. Bit timing (clocking)
Copyright C 1999 by Marko Vuskovic

1-11

OVERVIEW (Cont.)
Each protocol layer/sublayer adds its own header (and trailer) to PDUs coming from the upper layers and strips the according header/trailer from the PDUs coming from the south part of the protocol stack. Shown below is the general case of the protocol architecture. Not all adaptation sublayers are involved in a particular traffic type.

PDU from higher layers Header Higher layers Payload Trailer

SSCS Header header CPCS SSCS Header header header

Payload

Trailer

SSCS CS AAL CPCS SAR ATM TC PMD SAR header SAR trailer SAR header SAR trailer Payload Trailer CPCS trailer

SAR PDUs SAR Cell header header SAR trailer SAR Cell header header 48 octets SAR trailer

ATM Cells

Physical frames (DS-1, DS-3, STS-3,...)

Bit stream

11010101111010000111010010101111010101111010101110101010
Copyright C 1999 by Marko Vuskovic

1-12

OVERVIEW (Cont.)

Virtual Circuits ATM supports three types of switched connections: Permanent Virtual Circuits (PVC) Switched Virtual Circuits (SVC) Soft PVC (SPVC) (X.25 supports PVC and SVC) A PVC is programmed by the network administrator and doesn't require additional work in a communication session. If the connection fails it can't be reestablished (there are no alternate routes.) An SVC is established each time the connection is needed and terminated there after. Signaling (call setup/call clearing) is a complicated procedure, which can cause incompatibility among different vendors, but provides rerouting capabilities in case of the connection failure. A soft PVC is a compromise solution in which the connection between the DTEs and the network are a PVC, while the connection across the net-

ATM network

PVC

DTE

SVC

PVC

DTE

In any case, PVC, SVC, or SPVC, there is no routing during a communication session. Instead, the cells are switched by simple hardware implemented algorithms, which eliminates processing delay at ATM network nodes.

Copyright C 1999 by Marko Vuskovic

1-13

OVERVIEW (Cont.)
Guaranteed Quality of Service This is the third important characteristic of ATM. The combination of small fixed-size cells, statistical TDM and virtual curcuits provids a possibility of guaranteed delivery of a certain traffic type. The traffic type can be defined by the following negotiable QoS parameters: Cell Delay Variation (CDV) Cell Transfer Delay (CTD) Cell Loss Ratio (CLR) There are other QoS parameters which are not negotiable, but which characterize the traffic: PCR - Peak Cell Rate SCR - Sustainable Cell Rate MCR - Minimum Cell Rate CER - Cell Error Rate CMR - Cell Missinsertion Rate Required QoS parameters are normally specified by the application.
Small loss of speech is not critical, but delays can cause echo

10-4

Voice File transfer

10-6

Interactive data

Web browsing

10-8 Interactive video 10-10 Broadcast video Broadcas video can tolerate larger delays (playback buffers can compensate for CVD) Maximal CDV 0.1 1 10 100 1000 10000 [msec]

Video depends on amount of motion, resolution and immage size. More demanding than voice in terms of cell loss

Used from D. McDysan and D. Spohn: "ATM -Theory and Applications," McGraw-Hill, 1999

Copyright C 1999 by Marko Vuskovic

1-14

OVERVIEW (Cont.)
LAN Emulation How can an existing LAN, without changes in interfaces, drivers and applications, take advantage of ATM-based WANs (for example: campus backbone)? This is a problem of interoperability of legacy LANs, which was addressed by the ATM Forum. The solutions is in LAN Emulation, which consists of a set of standards that enable Ethernet (802.3) and token rings (802.5) to communicate across an ATM network, taking a full advantage of its benefits (low latencies, performance, scalability).
Legacy LAN (802.3)

LANs are unaware of existence of ATM WAN

ATM Backbone

B
Legacy LAN (802.5)

SVC

LEC
LAN Emulator Client (a router)

LEC
LAN Emulator Client (a router)

Host A
Application TCP/IP LLC LLC MAC PHY MAC frames MAC PHY ALL ATM PHY

LEC
Bridging LANE

Maps MAC/IP address (legacy address) into ATM address/SVC id. For that purpose an ARP server is needed, which runs on either LEC, or ATM switch.

ATM Switch
ATM PHY ATM cells ATM cells

Host B
Application TCP/IP LLC LLC MAC PHY MAC frames MAC PHY ALL ATM PHY ATM cells

LEC
Bridging LANE

ATM WAN
ATM Switch
ATM PHY ATM cells

Copyright C 1999 by Marko Vuskovic

1-15

OVERVIEW (Cont.)
Circuit Emulation Some legacy devices like classic PBX, video conferencing equipment, and terminal multiplexers are based on TDM circuits, which are all characterized as real-time communication systems with constant bit rate. Most of these devices use T1, E1, T3 or E3 circuits as private, or leased lines. The Circuit Emulations Service (CES) provides support for transporting TDM traffic across ATM networks, thus enabling the integration of legacy applications into a new ATM backbone and therefore protecting the existing investments into equipment.

CODEC T1/E1 PBX T1/E1 T1/E1

MUX

T1/E1 MUX

ATM Network

T1/E1 T1/E1 CODEC

PBX

CES Interworking Functions (Enable communication between TDM and ATM interfaces)
T1/E1 User

Ingress Switch
AAL CES-IWF ATM PHY

Egress Switch
AAL ATM PHY CES-IWF T1/E1 T1/E1 T1/E1 User

T1/E1

T1/E1

Native T1/E1 frames

ATM Network

Native T1/E1 frames

Copyright C 1999 by Marko Vuskovic

1-16

OVERVIEW (Cont.)
ATM Interfaces Most of the ATM protocols are groupped according to the interface types. There are three basic interfaces: UNI - User-Network Interface (Interface between ATM end users and a private ATM switch, or between a private ATM switch and the public carrier ATM network) PNNI - Private Network-Network Interface (Interface between two switches in the same ATM network) B-ICI - Broadband (ISDN) Inter-Carrier Interface (Interface between two public ATM networks)

NOTICE: Sometimes NNI is used instead of PNNI. Strictly, NNI denotes a generic term: "network node interface", interface between two network nodes. In addition to the basic interface, there are others like LUNI, FUNI, LNNI, etc., which will be discussed later.

Private Network

PBX

UNI UNI

Router

UNI

Public ATM Carrier Network

UNI
Private ATM switch

PNNI PNNI PNNI PNNI PNNI

B-ICI
Public ATM Carrier Network

Copyright C 1999 by Marko Vuskovic

1-17

PHYSICAL AND ATM LAYERS

Cell Header There are two slightly different versions of the ATM cell header, one for UNI and another for NNI. NNI doesn't use generic flow control (GFC), therefore it can use larger virtual path identifier. The goal of flow control at the local level is to work with the QoS and help decrease the congestion problem (as will be seen later, cells get dropped at congested nodes, which can make the traffic problem even worse due to retransmission requests made by higher protocol layers.) GFC is to prevent excessive retransmissions in a way similar to frame relay. The specification of GFC is not yet finished; the GFC field is normally not used (set to 0x0).

ATM Cell (53 octets) Header (5 octets) Payload (48 octets)

UNI
4 bits 8 bits 16 bits 3 bits 1bit 8 bits

GFC
Generic Flow Control

VPI

VCI
Virtual Channel Identifier

PT
Payload Type

HEC
Header Error Control

Virtual Path Identifier

Cell Loss Priority

NNI
12 bits 16 bits 3 bits 1bit 8 bits

VPI

VCI

PT

HEC

The header fields and their usage will be discussed in the following sections

Copyright C 1999 by Marko Vuskovic

1-18

PHYSICAL AND ATM LAYERS (Cont.)

Virtual Paths and Virtual Channels ATM uses virtual circuits (as in X.25 and FR). However, the virtual circuits are hierarchically organized using three concepts: Transmission Paths (TP) - Physical connections between network nodes Virtual Channel Connection (VCC) - Logical end-to-end connection between two users, which provides transport of a particular transaction (voice, video, data). Conceptually similar to virtual circuits in X.25. A VCC preserves the cell sequence integrity. VCCs are unidirectional. Virtual Path Connection (VPC) - A bundle of VCCs with identical routing. Introduced to organize the VCCs and to manage the network more efficiently. VCCs in a VPC can have different traffic parameters and differently negotiated QoS parameters. VPCs are unidirectional. A VCC is a concatenation of virtual channel links (VCL). Each VCL is labelled with an identifier called Virtual Channel Identifier (VCI). A VCP is a concatenation of virtual path links (VPL). Each VPL is labeled with an identifier which is called Virtual Path Identifier (VPI). VCIs and VPIs are used for routing and switching. Theoretically VCIs are in the range 0..65535, while VPIs are in range of 0..255 for UNI, and 0..4095 for NNI. (Practically not all switches support the full ranges). VCIs and VPIs have local significance, i.e. they are re-mapped in each network node. Reason: easier network reconfiguration and smaller numbers (i.e. less space in cell header). The knowledge of (VPI.VCI) pairs along a VCC is distributed across the network, and is known after the connection establishment.

15 Mbps (CBR)

64 kbps (CBR, occasional cell loss is OK)

Virtual Path Connection (VPC)

64 kbps (CBR) 15 Mbps (CBR)

128 Mbps (bursty, no cell loss)

Virtual Channel Connections (VCC)

Virtual Channel Connections (VCC)

128 Mbps (bursty)

Copyright C 1999 by Marko Vuskovic

1-19

PHYSICAL AND ATM LAYERS (Cont.)

Permanent Virtual Circuits VCCs and VPCs can be preestablished (PVC - Permanent Virtual Circuits), or set up on demand (SVC - Switched Virtual Circuits). In PVCs the routes are set manually once forever (until the next change) by the network administrator. SVCs are established and terminated by the call setup/call clearing procedures. One of the advantages of permanent virtual circuits is PVP tunneling: a public carrier can create permanent virtual paths (PVPs) across its public ATM network. These paths logically connect a couple of private ATM networks. The PVPs make the public ATM network transparent to the private ATM networks, i.e. several private ATM networks become integrated into a larger private ATM network. The user can establish his/her own VCCs through the PVP tunnels (even if the public carrier doesn't support VCs) - the condition is that the total bandwidth of the VCCs are within the bandwidth of the PVP.

Private ATM network

PVP Tunneling

Private ATM network

PVP PVP

Public ATM network

Private ATM network

Permanent Virtual Paths

Copyright C 1999 by Marko Vuskovic

1-20

PHYSICAL AND ATM LAYERS (Cont.)

ATM Switching There are three types of ATM switching: virtual path switching, virtual channel swiching, and combined virtual path/virtual channel switching. Virtual Path Switching Switching based upon VPIs only (the VCIs are just passed without modification). End-users may assign VCIs arbitrarily and this will not affect the network. VCIs 1 2 3 1 2 3

TP2
Port 2 Port 1 VPI=12 1 VPI=6 VPI=18 VPI=33 2 Port 3

1 VCIs 2 3 VCIs VCIs

1 2 VCIs

TP1
2 1 2

VPI=52 VPI=8

TP3

1 2 1

3 VCIs VCIs

ATM switch
Input Port VPI 1 1 2 3 12 52 24 33 Output Port VPI 2 3 3 1 61 18 6 8

Copyright C 1999 by Marko Vuskovic

1-21

PHYSICAL AND ATM LAYERS (Cont.)

Virtual Channel Switching VPIs are set to 0 and are left unchanged in each ATM switch, while only VCIs are remapped. Combined Virtual Path/Virtual Channel Switching Switch re-maps the entire label, VCI and VPI. The highest flexibility in switching.

ATM switch
VC switch

VCI=223 VCI=454 VCI=670 VCI=665

VPI=33 VPI=12 TP1 VPI=52 VPI=24 VPI=18 TP2

VCI=52 VCI=51 VCI=670 VCI=665

VP switch

Input Port VPI VCI 1 1 1 1 12 12 52 52 223 454 670 665

Output Port VPI VCI 2 2 2 2 33 24 18 18 52 51 670 665

Copyright C 1999 by Marko Vuskovic

1-22

PHYSICAL AND ATM LAYERS (Cont.)

Payload Type The most significant bit indicates user data or network management cells. This allows the insertion of management cells into a user's VCC without disrupting the user's data (inband control). The middle bit of user data cells is Explicit Forward Congestion Indicator (EFCI) (similar to FECN - Forward Explicit Congestion Notification in frame relay). EFCI can be set by a congested node, to notify the destination end system, which may implement an algorithm for adaptive lowering of the cell rate during congestion periods. The least significant bit of user data cells is used in ATM Adaptation layer.

PT
0 0 0 0 1 1 1 1 0 0 1 1 0 0 1 1 1 0 1 0 1 0 1 0 SDU type 0 SDU type 1 SDU type 0 SDU type 1 segment end-to-end

Congestion not experienced Congestion experienced OAM cell

User Data

Resource management Reserved for future

Network Information

SDU - Service Data Unit (refers to 48-octet payload) OAM - Operations, Administration, and Maintenance

Copyright C 1999 by Marko Vuskovic

1-23

PHYSICAL AND ATM LAYERS (Cont.)

Cell Loss Priority The CLP field defines priority when it comes to cell discard due to congestion. If CLP = 1 - cell gets low priority and will be discarded by the congested node if needed. In severely congested traffic, the cells marked CLP = 0 can be discarded either, but not before the low priority cells. The ingress network node can set CLP if the corresponding VCC violates the agreed traffic parameters. For example, certain VCC can exceed the cell rate which was negotiated at the connection establishment. The violating cells can still go through the network, but in case of congestion they will have a low priority (see later, chapter "Traffic Management") The end user can also set CLP. For example, user can insert cells that are beyond the negotiated cell rate with CLP = 1. The cell will be delivered to the destination if the traffic conditions in the network are convenient.

Violates the agreed cell rate

VCI=12 CLP=0 VCI=33 CLP=0 VCI=12 CLP=0 VCI=07 CLP=0

VCI=12 CLP=1 VCI=33 CLP=0 VCI=12 CLP=1 VCI=07 CLP=0 VCI=33 CLP=0 VCI=07 CLP=0

Router

Ingress switch

Congested switch

Discarded cells

Copyright C 1999 by Marko Vuskovic

1-24

PHYSICAL AND ATM LAYERS (Cont.)

Header Error Control The header field HEC protects only the header (not the payload!). Error control of the payload is performed by higher layer protocols and applications. The integrity of the header is extremely important because a corrupted header can cause delivery of cells to wrong addresses, or can interfere with network management. This can bring the network in a catastrophic condition.

HEC protects the entire header including the HEC itself

Rest of the header

HEC

Payload

Protected by HEC

Unprotected by HEC (protected by higher layers)

A study has shown that among all errors in fiber-optic transmission media, 99.64% of them are single bit errors. This means that two-bit or three-bit errors are not very likely. However, it is also known from the experience that the large error bursts have also a high probability (of course, far less than single bit errors, but far more than two, three or four bit errors). This fact has influenced the design of the ATM error control mechanism and the size of the HEC field. ATM layer uses the Bose-Chadhuri-Hocquenghem (BCH) algorithm which corrects singlebit errors and detects multiple bit errors. The algorithm is based on a combination of the 8-bit CRC scheme (which can detect errors) and Hamming code (which can correct single bit errors). This combination is based on the redundancy of HEC, which has 8 bits and is used to protect only 5x8 = 40 bits. The coding theory shows that only 6 bits are necessary to detect and correct a single bit error in a message that has 40 bits. Furthermore, an extended 8-bit HEC can correct single-bit errors and detect 84% of other errors. The algorithm computes the HEC value by modulo-two division of the header bit pattern (with the HEC field initialized to zeroes) by the CRC generator whose polynomial is x8 + x2 + x + 1. The remainder of the modulo-two division is then OR-ed with 01010101 and placed into the last octet of the cell header (HEC field). Similar operation is performed at the receiver's side, where the entire header (including HEC field) is modulo-two divided by the CRC generator. The error correction capability greatly diminishes the need for cell discarding, and consequently, the need for cell retransmission.
Copyright C 1999 by Marko Vuskovic

1-25

PHYSICAL AND ATM LAYERS (Cont.)

The working of the HEC algorithm (which uses BCH error coding) can be described by the following two-state diagram:

No error detected (pass the cell)

Multibit error detected (drop the cell) No error detected (pass the cell)

Error (any) detected (drop the cell)

Correction mode

Detection mode

Single-bit error detected (correct the header and pass the cell)

The algorithm starts in "correction mode" state. In detection mode the cells have an opportunity to be corrected (if the error is single-bit). If any error happens, the state is changed to detection mode, and it stays there as long as errors persist. This is designed to handle error bursts. It is assumed that the burst has stopped if a valid cell entered the switch - then the state is changed back to the correction mode.

10-4 10-8 10-12 10-16 10-20 10-24 10-28 10-10 10-9 10-8 10-7 10-6 10-5

BER = 10-6, p = 10-8 (UTP) BER = 10-10 , p = 10-17 (fiber optic)

Bit Error Rate

Used from: U. Black: "ATM Foundation for Broadband Networks", Vol. 1, Prentice Hall, 1999 (Originally published by ITU-T in I.432)

Copyright C 1999 by Marko Vuskovic

1-26

PHYSICAL AND ATM LAYERS (Cont.)

Cell Alignment Since cells have no flags to indicate their beginning and the end, there must be some "pattern recognition" algorithm to find the exact location of a cell in a stream of incoming bits. To accomplish this, the receiver checks if a candidate header is valid: it modulo-two divides the 40 bits of the supposed header with the CRC generator. A zero remainder would indicate the possibility for a valid header. If the remainder is not zero, the receiver moves one bit further and perform the check again. This process continues, bit-by-bit, until a successful match occurs. As seen, the HEC has two functions: error control and cell synchronization (alignment, delineation). Supposed to be header (apply HEC) 40 bits

NO MATCH

10101000101000111100001011101010011100011100010101010101010101010
Payload HEC Rest of the header Payload

...

...

... ...

Supposed to be header (apply HEC) NO MATCH 40 bits

10101000101000111100001011101010011100011100010101010101010101010
Payload HEC Payload

...

...

... ...

Supposed to be header (apply HEC) MATCH ! 40 bits

10101000101000111100001011101010011100011100010101010101010101010
Payload HEC Payload

...

...

... ...

Copyright C 1999 by Marko Vuskovic

1-27

PHYSICAL AND ATM LAYERS (Cont.)

Cell Scrambling What happens if a payload accidentally contains a 40-bit subpattern which satisfies the HEC? This situation must be not permitted. Therefore the ATM forum has proposed an optional capability of cell scrambling, which looks for a bit pattern within the payload that satisfies the HEC. If such bit pattern is found, it is modified so that the match cannot occur within the payload, and that the payload can be easily unscrambled at the receiver's side.

Repetitive Checking The HEC is not 100% accurate. Therefore the check must be performed several times successfully on three consecutive cells before the receiver finally can decide that it has found the right header and is in synchronism. This is called repetitive checking. A usuall value of repetitions is three times.

3rd successful HEC match

Jump 48 octets and check again

Jump 48 octets and check again

1st successful HEC match

This is a valid cell!

2 nd successful HEC match

Copyright C 1999 by Marko Vuskovic

1-28

PHYSICAL AND ATM LAYERS (Cont.)

The repetitive checking algorithm, which is performed at the receiver's side, can be represented as a three-state diagram. The algorithm starts in "hunt mode" where it moves bitby-bit in search for a valid header. Then it goes to "presync mode" where it continues to move cell-by-cell, looking for three (in general case ) successful header matches, which is a condition for synchronism. In "sync mode" the algorithm starts regular error checking based on the HEC field. If there are more than unsuccessful HEC, the receiver considers itself to be out of synchronism, and goes back to "hunt mode".

Successful match Unsuccessful match (move bitwise)

HUNT

Unsuccessful match

PRE SYNC

Successful match (move times, cellwise)

th unsuccessful match

th successful match

SYNC

Receiver enters correction mode and performs regular HEC (move cellwise)

Copyright C 1999 by Marko Vuskovic

1-29

ADAPTATION LAYER
The purpose of the ATM adaptation layer (AAL) is to adapt (converge) the user traffic (applications) to a cell-based network. The user traffic can have various characteristics and various demands regarding the bandwidth, burstiness, delay tolerance and cell loss tolerance. Therefore there are several different AAL stacks, labeled AAL1, AAL2, AAL3/4 and AAL5. In order to understand the differences between various AAL types, we must define first the existing service categories and the traffic classes. Service Categories (Here will be given the basic definitions. More details will be discussed in chapter "Traffic Management"). There are five service categories defined as follows:

CBR - Constant Bit Rate This is the most expensive service, which gives a guaranteed amount of bandwidth to a VC. There is no elasticity in the usage of bandwidth. In that it is similar to a leased T-1 or T-3 line. It is highly inefficient if it is not used 100% of time. Used for real-time applications that are very sensitive to transfer delay and cell delay variation, such as voice, video and circuit emulation.

Time The following applications mostly use CBR: Interactive video (video conferencing) Interactive audio (telephone call) Video distribution (TV, distributed classroom) Audio distribution (radio, audio feed) Video retrieval (video on demand) Audio retrieval (audio library)

Copyright C 1999 by Marko Vuskovic

1-30

ADAPTATION LAYER (Cont.)

rt-VBR - Real-Time Variable Bit Rate Supports real-time applications that are bursty in nature and therefore can allow more efficient use of a network. The transfer delay (CTD) and delay variation (CDV) are tightly controlled. Examples are voice coders with compression and silence suppression, and other applications that generate variable frame sizes.

PCR MBS SCR

Time SCR - Sustainable Cell Rate (upper edge of average cell rate) PCR - Pick Cell Rate (Upper limit that the application should never exceed) MBS - Maximal Burst Size (number of cells which exceed SCR, and which can be tolerated)

The network allocates resources (bandwidth) according to SCR. The cells that exceed PCR will be flagged by CLP after MBS. Those cells violate the traffic contract and will be drooped in case of congestion (see subsection "Cell Loss Priority" in the previous chapter.) nrt-VBR - Non real-Time Variable Bit Rate Used for non-real time applications which are more tolerant of network delays. CDV is not controlled, cell loss is controlled. Response time critical transaction processing, such as: airline reservations banking transactions process monitoring store and forward video
Copyright C 1999 by Marko Vuskovic

1-31

ADAPTATION LAYER (Cont.)

ABR - Available Bit Rate Designed for non real-time applications which tolerate network delays and cell loss (such as LAN interconnect and Internet traffic) - a best effort service. Allows most of the statistical gain without guarantee for throughput. Allows congestion control because the sender can be informed (requested) to slow down the traffic in congestion periods. ABR provides a very good network utilization.

PCR

MCR

Time PCR - Pick Cell Rate (Upper limit that the application should never exceed) MCR - Minimal Cell Rate (can be 0)

Typical applications: Critical Data Transfer (defense information) Interactive Text/Data/Image Transfer (banking transactions, e-mail, telex, fax) Text/Data/Image Distribution (newsfeed, weather satellite pictures) Test/Data/Image Retrieval (file transfer, library browsing) Aggregate LAN (LAN interconnection, LAN emulation) Remote Terminal (telnet, telecommuting)

Copyright C 1999 by Marko Vuskovic

1-32

ADAPTATION LAYER (Cont.)

UBR - Unspecified Bit Rate Best effort service without any performance requirement. Only the PCR is specified (which normally equals to the line bit rate). UBR is equivalent to Internet.

PCR

Time

The service categories can be summarized as follows:

Service Category CBR rt-VBR nrt-VBR ABR UBR

Network Priority 1 2 3 4 5

CTD, CDV low low high high high

Cell Loss low med med med high

Burstyness none small small high high

Bandwidth Guarantee yes yes yes optional no

Copyright C 1999 by Marko Vuskovic

1-33

ADAPTATION LAYER (Cont.)

Traffic Classes The various AAL types are defined according to traffic types, which in turn are classified with respect to the following three attributes: the timing relationship required between the source and destination, the bit rate, and the connection mode. These summarized in the following table:

Traffic Class Timing relation between source and destination Bit Rate

Class A

Class B

Class C

Class D

Required

Not required

CBR

VBR

Connection Mode

Connection-Oriented

Connectionless

AAL Type

AAL1

AAL2

AAL3/4 or AAL5

Example Application

T-1, E-1 circuit emlation

Packet video, audio

FR, X.25

IP, SMDS

Copyright C 1999 by Marko Vuskovic

1-34

ADAPTATION LAYER (Cont.)

Adaptation Layer 1 (AAL1) Generally designed for CBR traffic (uncompressed real time voice and video). Specifically it provides circuit emulation of full or fractional DS-1/E-1. There are two formats of CES: Unstructured CES Structured CES (Fractional CES) Unstructured CES, also called "clear channel", takes the entire TDM frame (T1 or E1) and passes is to SAR as an integral PDU. Along with the 24 (or 30) 64 kbps channels the signalling bits are carried through the network as well. Structured CES, also called "channelized T1 or E1", dissasembles the T1 or E1 frame and treats individual 64 kbps channels, or their combinations, as independent entities, which can be routed to different destinations. The signaling bits can not be carried as in unstructured CES, they are added to the end of the emulated fractional data stream.
12 x 64 1 x 64 4 x 64

ATM Switch

1 x 64 2 x 64

4 x 64 12 x 64

ATM Switch

ATM Switch

2 x 64 12 x 64

6 x 64

6 x 64

12 x 64

Copyright C 1999 by Marko Vuskovic

1-35

ADAPTATION LAYER (Cont.)

Protocol stack for unstructured service

DS-1 (or E-1) bit stream F F F F

CES Interworking Function (CES-IWF) maps CBR stream into 47-octet PDUs, without being aware of the frame structure (it doesn't see time slots, channels, D0s). The mapping implicitly includes signaling and framing bits.

CS
47-octet CS-PDU 47-octet CS-PDU 47-octet CS-PDU

The receiving CS handles the cell delay variation and delivers the CS-PDUs to the application at the constant bit rate. Detects lost or missequenced cells Provides Forward Error Control (FEC) on the AAL1 header and optionally on data Provides source clock frequency recovery at the receiver The transmitting CS passes AAL1 header data to SAR sublayer Passes CS-PDUs to SAR sublayer

SAR
1-octet AAL1 header 3 Sequence CSI Count 1 Sequence Number 3 1 47 - octet payload Sequence count is used to detect lost or out of sequence cells. CRC is used for single error correction and multiple error detection in first 4 bits of SAR header.

CRC

Sequence Number Protection

Even parity bit is used for protection of CS Indication the first 7 bits of SAR header. (Spans several frames. In odd cells carries timing recovery info, in even cells contains an indication of structured or unstructured service.)

Copyright C 1999 by Marko Vuskovic

1-36

Copyright C 1999 by Marko Vuskovic

1-37

ADAPTATION LAYER (Cont.)

Adaptation Layer 2 (AAL2) Designed for real-time VBR low bandwidth compressed voice (ADPCM, CELP, ACELP) and video traffic (MPEG-2, JPEG, H.320). AAL2 has strong point in the trunking of multiple (compressed) voice channels over a single VCC. This standard is relatively new (recommendation proposed by ITU-T in 1997). Before AAL2, for compressed video and voice was used either AAL1 or AAL5. AAL2 doesn't have SAR sublayer. VBR Sources

CPCS

CPCS packet header CID 8 bits User 1 LI 6 UUI 5 HEC 5 User 2 Pk hdr 3 octets Data Packet 1-64 octets Pk hdr 3 octets User 3 Data Packet 1-64 octets

Pk hdr 3 octets

Data Packet 1-64 octets

CPCS hdr 1 octet 6 bits OSF 1 SN 1 P

CPCS Payload 47 octets

CPCS hdr 1 octet

CPCS Payload 47 octets

Cell Header

ATM Payload

CID - Channel Identifier (needed to multiplex several users onto a single VCC) LI - Length Indicator (needed because the packets have variable length) UUI - User to User Indication (to identify a particular SSCS layer) HEC - Header Error Control (protects CPCS packet header) OSF - Offset Field (Identifies the next CPCS packet header within the CPCS payload) SN - Sequence Number (Because packets can span max two cells, sequence numbers are 0,1) P - Parity Bit (protects/corrects SN)
Copyright C 1999 by Marko Vuskovic

1-38

ADAPTATION LAYER (Cont.)

Adaptation Layer 3/4 (AAL 3/4) Adaptation layers 3 and 4 were originally designed as separate layers, but were later combined into a single layer. AAL 32/4 is generally designed for all type of traffic that is not real-time critical. AAL 3/4 uses CPCS and SAR sublayers.

Application Data Packet (1-65,535 octets)

CPCS
1 CPI 1 Btag 2 octets BAsize Makes CPCS payload multiple of 32 bits CPCS Header 1/2 octet CPCS Payload Padding 0-3 octets CPCS Trailer 1/2 octet 1 AL 1 Etag 2 octets Length

SAR
2 44 octets 2 2 44 octets 2 2 44 octets 2

SAR SAR SAR Payload Trailer Header

SAR SAR SAR Payload Trailer Header

SAR SAR SAR Payload Trailer Header

ST 2

SN 4

MID 10 bits

LI 6

CRC 10 bits

Segmentation/reasembly Error detection/control Sequence integrity Multiplexing (interleaving)

CPI - Common Part Indicator (unit of measurement for BAsize, bits or octets) BAsize - Buffer Allocation Size (size of the CPCS payload) Btag, Etag - Identical bit patterns, compared at receiving side, if different cell is discarded) AL - Alignment field (makes the trailer a full 32 bits field to simplify the receiver design) Length - length of the CPCS PDU (difference is in padding) ST - Segment Type (BOM = beginning of message, COM = cont. of mes., EOM = end of message, SSM = single segment message) SN - Sequence Number (used to detect lost SAR PDUs) MID - Multiplex ID (allows multiplexing of up to 1024 different CPCS PDUs over the same VCC) LI - Length Indicator (= 44 for BOM or COM, can be less than 44 in EOM, SSM) CRC - Protects complete SAR PDU, including header, payload and trailer.
Copyright C 1999 by Marko Vuskovic

1-39

Copyright C 1999 by Marko Vuskovic

1-40

Copyright C 1999 by Marko Vuskovic

1-41

ADAPTATION LAYER (Cont.)

Adaptation Layer 5 (AAL 5) AAL 3/4 has too much overhead. AAL 5 was originally designed by IBM, later accepted by ATM Forum and ITU-T, to make a more efficient adaptation that can support TCP/IP, which is already self-sufficient and doesn't need extensive error/sequence control of AAL 3/4. Later was added support to other protocols, LANs, FR, VBR and CBR voice and video. CPCS sublayer has only a trailer, while SAR sublayer doesn't append any overhead. It only uses PT (payload Type) from the ATM cell header, to indicate when the reassembly can begin (its caled AAL_indicate). Application (voice, video, TCP/IP/LLC, FR,...)

CPCS
PAD UU CPI Length CRC

0-47 octets

CPCS Payload (1..65,535 octets)

CPCS Trailer

SAR

SAR Payload (48 octets)

SAR Payload (48 octets)

SAR Payload (48 octets)

Cell Header PT ATM Cell Payload (48 octets) AAL_indicate (lsb of PT - 1 if the last cell in a PDU, otherwise 0) PAD - Padding (to make the payload a multiple of 48 octets) UU - User-to-User Indication CPI - Common Part Indicator (unit of measurement for the length field, bits or octets, also aligns the trailer to a 64-bit boundary) Length - The length of the CPCS payload (without padding) CRC - CRC-32 (Error detection of the CPCS PDU, not very reliable for long data packets, same as in 802.2)
Copyright C 1999 by Marko Vuskovic