Asynchronous transfer mode (ATM).

ATM has been universally declared as the standard for B-ISDN. It was developed independently in the United States by Bellcore and standardized through ANSI, and in Europe where it was standardized through ETSI. The small differences between the two a gradually converging. The major objective of ATM is to integrate real-tin information such as voice and video with non-real-time computer data, with the same transmission and switching medium. The conflicting require for these two categories can basically be stated as follows:

• · Data requires very low BER but can tolerate large propagation delays (seconds).
• · Voice and video require small propagation delays (milliseconds) but can tolerate some errors or small losses of information.

Chronological SDH penetration.
(a) Fiber from trunks to feeder.
(b) fiber into distribution loops.
(c) fiber to the home.

These differences can be observed as the different sampling or scanning times. For voice, regular 8000 samples per second are necessary, whereas video might need a short burst of several megabytes followed by nothing for a few seconds. In contrast, computer data might need a high data rate for some applications, such as image transfer, but a low data rate for simple text transfer. TDM processes such as PCM are considered inefficient because time slots are allocated whether or not information is placed in them. PCM and therefore N-ISDN are examples of the synchronous transfer mode (STM).

In ITU-T Recommendation 1.121, the asynchronous transfer mode (ATM) is used for the ISDN user network interface. The Recommendation states that although B-ISDN will support circuit mode applications, it will have a packet-based transport mechanism. That means N-ISDN will evolve from a circuit-switched telephone network to a packet-switched B-ISDN. ATM has some similarities to and some major differences from X.25. One difference is that ATM uses common channel signaling, whereas X.25 has control signaling on the same channel as the data transfer.

ATM improves on the inefficiencies of STM by packetizing the data into cells. Every cell is given a connection address and sent on a fixed route through the network. Packets are made small enough so that if they get lost or collide with other packets sharing the same route, they can be resent if necessary.

Algorithms are used to compensate for the variable delay inherent in packet transmission so that ATM can be used for continuous flow applications such as telephony and video. ATM has considerable flexibility in its use for all services. New services can be added without restructuring the network, or switching exchanges and services already in existence can be modified or expanded. For example:

• Broadband unrestricted bearer services
• High-quality broadband video telephony
• High-quality broadband videoconferencing
• Existing quality TV distribution
• HDTV distribution

ATM networks

ATM is a connection-oriented technique and ATM networks are essentially fast packet-switching networks. Each packet has a fixed length and is called a cell to distinguish it from its X.25 equivalent, which is a variable length packet. The ATM destination address contains a virtual circuit identifier (VCI) or virtual path identifier (VPI) carried in the fast packet header instead of the time slot used in the STM systems. The VCI or VPI identifies a virtual channel such that in a virtual connection all cells of the same connection are transferred through the same route. This ensures cells are received in the same order as transmitted. The VCI is established during the call setup and released on completion of the call. Signaling and information are transported on separate virtual channels.

The unused time slot inefficiency of PCM is overcome by the fast packet-switching process called statistical multiplexing. Several connections are made over the same link depending on their traffic details. Voice coexists with data by sending bursts of data during the inactive voice times. If simultaneous bursts are transmitted, buffering enables the voice to take priority while the data is held back for transmission possibly a few milliseconds later. The statistical multiplexing process evens out the different types of traffic over time to fill the transmission link. This process is not possible with the more rigid formatting of PCM in the synchronous transfer mode. Interestingly, an ATM system that is very heavily loaded with traffic degenerates into an STM-like system. Congestion is currently a hot topic in ATM research.

An important question to answer is, How will the ATM network exist with the present LAN networks such as Ethernet, token rings, or FDDI? As with all new technology, there must be a transition phase, or overlay period, during which the old and the new inevitably coexist. Users gain access to the ATM network through a user network interface (UNI). In the case of the ATM connection to a standard Ethernet or token ring LAN, the UNI converts the LAN frames into ATM cells on entry and ATM cells to LAN frames on exit. The heavy investment in LAN technology dictates a gradual transition to ATM networks. The full power of the ATM network will be experienced only when telephones, videos, and computer terminals (DCE) connect directly to the ATM network through an ATM interface. Here, the ATM interface statistically multiplexes the different media of voice, video, and data and becomes the UNI for transmission over the ATM backbone network.

ATM cells.

ANSI has defined each ATM cell to have 53 bytes as shown in Fig. 10.36. The header has 5 bytes and contains the VCI label (24 bits), control bits (8 bits), and checksum (8 bits). The rest of the cell (48 bytes) is for data, which can be partitioned, if desired, into a 4-byte adaptation layer and 44 bytes of data. The adaptation layer allows the flexibility to subdivide and reassemble cells prior to transmission and after reception. One of the 8 bits of header control is used to set data at 44 or 48 bytes. At certain intervals, the control bits will be used to identify a cell used as a flow control cell that does not contain , customer data. The control also contains 1 bit to indicate that the cell can be I deleted in situations of extreme congestion. The ETSI ATM cell is also defined as 53 bytes long, containing 48 bytes of data and 5 bytes of header. The differences lie in the composition of the VCI and the checksum fields.

ATM network connections.

When a user wants to access the ATM network, the UNI is contacted first. The UNI is given information about the type of traffic, destination, bit rate, cell loss acceptability, and delay acceptability, and forwards these details to the network. The network uses these parameters to establish the best available route for transmission. A connect setup request is sent to all nodes in the proposed route. A virtual connection is made right through to the destination.

Remember, a virtual connection is one that is made only for the duration of the information transfer. Data is then transferred at the prenegotiated quality of service. On termination of the virtual connection the VCI is reallocated to another connection. This virtual circuit connection differs from that of the X.25 packet-switching process in that ATM has no acknowledgment sent back from the destination to source. This is a speed advantage at the expense of arrival verification.

Also, ATM cells are not sequence numbered like X.25 packets. When a virtual circuit has been set up, there is only one possible path for cells to take, so they cannot be received in the wrong order. Cells might be buffered at some nodes to assist with flow control, but they still arrive at the destination in the same order in which they were transmitted. If cells are sent into the network using the adaptation layer without prearranging a virtual circuit, they can arrive at the destination in the wrong sequence. This adaptation layer requires a sequence number to bf allocated to ensure correct reassembly at the destination.

ATM Standard.

ATM can be related to the OSI scheme. The physical layer has an SDH lowest bit rate of 155 Mb/s (STS-3c or OC-3) for ATM traffic. Other SDH rates can be used, such as 622 (OC-12) or 2.488 (OC-48). The ATM cell can be considered part of the data link layer. It also contains some network layer activities (adaptation layer) such as route setup, connection formation, and data flow control.

The transmission control protocol/Internet protocol (TCP/IP) can interact with ATM even though the IP is a connectionless protocol. This is done by the fragmentation of IP packets into ATM cells using software at the UNI source and destination. In this manner, two LAN gateways can have a virtual connection to the ATM network.

In order to maximize throughput, ATM does not perform end-to-end flow control, but monitors the queues forming inside ATM switches at all nodes. In this manner, there is very early detection of congestion buildup. The streams causing the queue buildup are identified and flow control messages are sent to the UNIs causing the problem to reduce the rate, but not completely stop the input of data into the network. The outcome of maximizing throughput and not allowing requests for repeats or receipt acknowledgments is that there is no data arrival verification and, consequently, no quality of service check. However, the TCP/IP or higher protocol layers do have these higher functions.

Introduction to Voice-Over IP

Voice-Over IP is a transmission technique that for many organizations can make both Internet telephony and telephony over the Internet a practical reality. That transmission technique involves taking the advantage of the use of Asynchronous Transfer Mode (ATM) in a wide area network infrastructure. Because ATM supports multiple service types with varying levels of service guarantees, it becomes possible with the applicable hardware and software to take advantage of appropriate ATM classes of service to obtain the predictability and reliability required to transport voice end to end and obtain a high quality of reconstructed voice at the destination.

We will examine the format of ATM cells, the manner by which ATM switches operate, ATM transmission services, metrics, admission control into an ATM network, the ATM protocol stack and the different ATM classes of service that are tailored to support different applications. Concerning the latter, we will note how certain classes of service are more suitable for transporting real-time voice than other classes of service. Once the preceding is accomplished we will then examine how this technology can provide us with the ability to obtain a true Quality of Service when linking geographically separated IP and frame relay networks.

OVERVIEW

ATM represents a fast packet-oriented, switch-based transfer technology based upon asynchronous time division multiplexing. Instead of variable packets ATM uses fixed-length cells. Each ATM cell is 53 bytes in length consisting of a 48-byte payload and a five-byte cell header.

The use of a fixed-length cell and a header where routing information is contained enables ATM switching to be performed in hardware. While this fact may appear to be trivial it permits extremely fast cell switching to occur and allows ATM to be scaled from a Tl access line rate of 1.544 Mbps to an OC-48 rate of 2.5 Gbps.

Cell Formats

There are two ATM cell formats, with a slight difference between the two. Once cell format is used at the user-to-network interface (UNI). The second cell format occurs at the network-to-network interface (NNI).

The UNI represents the interface between a subscriber and an ATM network. In comparison, the
NNI represents the interface between two ATM networks.

As we can note by comparing the two cell formats/ the only difference between the two concerns the use of the first byte in each cell. We will compare and contrast the fields of both cell formats by discussing them in their order of placement in each cell.

Generic Flow Control

In the UNI cell header the first four bits represent the Generic Flow Control (GFC) field. This field can be used to define the parameters for flow control from a subscriber into an ATM network. Because flow control is not applicable when connecting ATM networks, the FGC field is not used in cells at the NNI.

Virtual Path Identifier

The Virtual Path Identifier (VPI) represents a unique identifier that denotes a virtual path on a physical circuit. The UNI cell format uses 8 bits/ which provide 256 unique values.

In comparison, the NNI ATM cell format is extended to 12 bits through the use of the GFC field. This results in the NNI cell format providing 4096 possible values.

This structure enables ATM switches to rapidly route cells based upon their VIP field value within an ATM network. Once we complete our examination of the fields in an ATM cell we will note an example of cell routing in an ATM network, which will hopefully clarify the use of VIPs and VCIs.

Payload Type Identifier

The Payload Type Identifier (PTI) is used to indicate the type of data transported in the pay-load portion of the cell.

Cell Loss Priority

The Cell Loss Priority (CLP) one-bit field is used to prioritize cells. When this field is set to a value of binary 1 it indicates a low priority, while when set to a value of binary zero it means the cell has a high priority.

When a switch is under congestion the CLP value is used to determine the sequence of cells to discard. That is the switch will first attempt to discard cells with the CLP set to 1 before discarding those with a value of zero.

Header Error Control

The Header Error Control (HEC) is 8 bits in length. This field is used to provide an error correction capability that can correct one-bit error in the cell header. The HEC field is only applicable to the ATM header and does not provide protection to the payload.

Services and Connections

ATM provides three types of transmission services. Those services include permanent virtual connections (PVCs), switched virtual connections (SVCs)/ and a connectionless service.

PVC

A permanent virtual connection is similar to a leased line. That is/ a PVC emulates a leased-line type of service, although it does not represent a physical circuit. Instead/ a PVC is created by using shared resources with other ATM users. Because the path from source to destination is dedicated, no call setup signaling is required.

svc

A switched virtual connection represents a temporary path established through an ATM network. This results in a need for signaling information to establish a call as well as to remove the virtual channel from service once the call is completed.

Connectionless

A third type of ATM service is connectionless service. Connectionless service is used for supporting LAN-to-LAN communications via an ATM network. Now that we have an appreciation for ATM services and connections, let's turn our attention to the flow of data through an ATM network.

Payload Value	Meaning
000	User data cell, no congestion, cell type 0
001	User data cell, no congestion, cell type 1
010	User data cell, congestion experienced, cell type 0
Oil	User data cell/ congestion experienced, cell type 1
100	Maintenance information between adjacent switches
101	Maintenance information between source and destination
110	Resource management cell
111	Reserved for future function
	Values of the ATM Cell PTI Field

Data Flow

Let's assume three data sources are assigned VCI values of 2, 3 and 4. Let's further assume VCI2 and 3 are to be transported via the ATM network from location A to location B while VCI 4 is to be forwarded to location C. Because they are being routed to a common location they can be transported via the use of a common VPI. VPI 8 was used to bundle the three VCIs from A to B. At location B we will assume VCI 50 is to be routed to location C. Because VCI4 will be switched at location B to location C, a new bundle would be created. In this example VPI 50 is shown as the path between locations B and C/ which transports VCI 4 and VCI 50 from B to C. This miniature example of ATM data flow also indicates that the VPI changes at each connection point in an ATM network.

ATM Metrics

There are several metrics that define the performance of ATM. Those metrics include the peak cell rate (PCR)/ sustained cell rate (SCR)/ maximum burst size (MBS)/ and variable bit rate (VBR).

Peak Cell Rate

The peak cell rate defines the upper boundary of an ATM connection or the maximum cell rate. The PCR is the inverse of the minimum interarrival time between cells.

Sustainable Cell Rate

The sustainable cell rate (SCR) defines the amount of traffic an end point can burst into a network. For those persons familiar with frame relay, the ATM SCR is similar to frame relay's Committed Information Rate (CIR).

Maximum Burst Size

The maximum burst size (MBS) represents the maximum number of cells accepted over a period of time. When the cell rate exceeds the MBS/ cells can be dropped.

Variable Bit Rate

There are two types of variable bit rates supported by ATM—variable bit rate real time (VBR-rt) and variable bit rate non-real time (VBR-nrt). The VBR represents traffic guaranteed for delivery under the SCR. However/ traffic above the SCR may be discarded if it exceeds the MBS. The VBR cannot exceed the PCR as the latter represents the port speed on an ATM switch.

ATM Connection and Admission

There are two classes of parameters that govern connection admission control into an ATM network. First, source traffic characteristics in the form of the peak cell rate, average cell rate, maximum cell burst and peak duration permitted are exchanged. Next, information concerning the required Quality of Service, such as cell transfer delay, delay jitter, cell loss ratio, and maximum cell burst and cell loss are exchanged. In addition, the cell loss priority bit in the cell header allows users to generate two different priority traffic flows. The resulting two classes are treated separately by the network connection admission control. Also note that the cell transfer delay (CTD) represents another important metric. The CTD represents the elapsed time between a cell exiting one measurement point and entering another measurement point.

Switch Performance

In addition to network performance parameters there are five parameters that characterize ATM switching performance. Those parameters include throughput, connection blocking probability, cell loss probability, switching delay, and cell delay variation.

The throughput is the rate at which cells depart a switch per unit time. The connection blocking probability represents the probability that demand exceeds resources available between ingress and egress ports on a switch. At this time cells will be dropped based upon the setting of the CLP bit in the cell header.

The cell loss probability represents the probability that a cell will be lost. This occurs when a switch queue is filled and additional cells arrive.

The switching delay represents the time required to route an ATM cell through a switch while the cell delay variation represents the probability that the delay of a switch exceeds a certain value. This metric is also referred to as jitter delay. While switch metrics are important when purchasing hardware it is the previously mentioned two classes of parameters that govern connection admission control and primarily govern the flow of traffic through an ATM network.

THE ATM PROTOCOL STACK

The ATM protocol stack is relatively simple, consisting of either two or three layers based upon the location of data flowing into/ through, and out of an ATM network. At end points where data enters or exits an ATM network, the protocol stack has three layers. Those layers are an ATM Adaption Layer (AAL) responsible for adapting different classes of traffic to the ATM layer, in effect taking protocol data units (PDUs) and breaking them into cells; an ATM Layer that is responsible for relaying cells between the AAL and the physical layer; and the physical layer. Because ATM switches only work upon cells, there is no need for a mechanism to convert PDUs into cells. Thus, the AAL is only applicable for ATM end points. Because the AAL is critical for providing a specific class of service, we will focus our attention upon the top layer in the ATM protocol stack.

Types of AAL

Originally, five types of AALs were recommended by the Consultative Committee for International Telephone and Telegraph (CCITT)/ the predecessor to the ITU Telecommunications Standardization (TS) body. Those AAL layers were referenced as AAL1 through AALS. However, layers 3 and 4 were merged into a common joint layer referred to as 3/4.

Sublayers

Each AAL consists of two sublayers, referred to as the Convergence Sublayer (CS) and the Segmentation and Reassembly (SAR) Sublayer.

Convergence Sublayer
The Convergence Sublayer (CS) supports end-user applications and is further subdivided into two sublayers—the Common Part Convergence Sublayer (CPCS) that is required and an optional Service Specific Convergence Sublayer (SSCS).

The Convergence Sublayer is responsible for managing multiple data flows between applications and the Segmentation and Reassembly Sublayer.

Segmentation and Reassembly Sublayer
The Segmentation and Reassembly (SAR) Sublayer is responsible for taking a PDU/ such as an IP datagram, and breaking it into a sequence of cells, as well as reassembling cells into PDUs.

The Convergence Sublayer provides the AAL service at the AAL-SAP (Service Access Point) for higher layers and is service-dependent. The SAP can be considered to represent a mailbox where data is sent to higher layers or received from higher layers. Now that we have an appreciation for how the AAL is subdivided, let's turn our attention to each ATM AAL. Because each AAL employs a specific

SAR and CS/ each supports a specific type of traffic, commonly referred to as a Class of Service. Thus, let's examine the four types of AALs and their relationship to different Classes of Service.

There are four ATM AALs/ numbered AAL1/ AAL2/ AAL3/4/ and AALS. As previously explained, two AALs were merged and are now referred to as AAL3/4.

AAL1

AALl provides a timing relationship between source and destination. This permits AAL1 to support connection-oriented services that require constant bit rates and have specific timing and delay requirements/ such as emulating a DSO or DSl.The ability to support a constant bit rate provides a true QoS capability. This is because CBR results in the output or egress of a bit stream that has an extract or near-extract relationship to the timing of the bit stream upon entering the ATM network.

AAL2

AAL2 provides more efficient support for voice over ATM. This AAL layer supports variable length packets within the ATM payload. Previously referred to as Composite ATM/ AAL2 is now referred to as the ITU-TS 1.363.2 standard.

AAL3/4

AAL3/4 supports both connectionless and connection-oriented variable bit rate services. However/ AAL3/4 has high overhead and is rarely supported.

AAL5

AAL5 supports connection-oriented variable bit rate services. In comparison to AAL3/4/ AALS provides a reduction in overhead and is popularly used for variable bit rate applications.

PDU to Cell Conversion

Each AAL has PDUs converted into cells through the use of the Convergence Sublayer and the Segmentation and Reassembly sublayer. However/ types 3/4 and 5 have their CS subdivided into Service Specific Convergence Sublayer and Common Part Convergence Sublayer (CPCS).

The SSCS is applicable to variable bit rate applications that require specific services. Now that we have a basic understanding of AALs, let's turn our attention to ATM Classes of Service.

ATM Classes of Service

Under ATM/ multiple traffic classes that are commonly referred to as service types are supported. Each traffic class or service type has a predefined characteristic as well as a level of service guarantee. Each traffic class definition is based on the use of three attributes: the timing relationship between the source and destination, the variability of the bit rate, and its connection mode.

Timing Relationship

The timing relationship between the source and destination defines the ability of the receiver to receive the original data stream at the same rate at which it was originated. For example, a voice conversation digitized at 64 Kbps via PCM must be "read" by the receiver at that data rate to be correctly interpreted. In comparison, a file transfer occurring via a Tl line into the Internet at a data rate of 1.544 Mbps could be correctly received via an egress access line operating at 56 Kbps. Although the reception of the file requires additional time, different transmission and reception rates do not inhibit the actual data transfer.

Bit Rate Variability

The second attribute governing the class of traffic is the bit rate. Some applications, such as digitized real-time voice, require a constant bit rate. Other applications, such as a file transfer, can occur successfully with either a constant or a variable bit rate.

Connection Mode

A third attribute governing the class of traffic is its connection mode. The connection mode can be either connection-oriented or connectionless. Connection-oriented means a connection must be established prior to actual data transfer occurring, while connectionless references transmission occurring on a best-effort basis, with an acknowledgment flowing back only after transmission was initiated. Examples of connection-oriented applications include voice calls and IBM SNA data sessions. Examples of connectionless applications include Ethernet transmission and applications that use UDP.

ATM's Class A is also commonly referred to as a constant bit rate (CBR) class of service. CBR is well suited for transporting digitized voice and in fact was designed primarily for supporting voice communications. CBR allows the amount of bandwidth, end-to-end delay, and the delay variation to be specified during call setup. Class B traffic is commonly referred to as variable bit rate real time (VBR-rt) as it requires a timing relationship. In comparison, because Class C traffic does not require a timing relationship, it is commonly referred to as variable bit rate non-real time (VBR-nrt). Finally, Class D traffic is commonly referred to as unspecified bit rate (UBR).

Class of Service	A	B	C	D
AAL Service	AAL1	AAL2	AAL5	AAL3/4
Timing (bit rate)	Constant	Variable	Available	Unspecified
Traffic	Delay sensitive	Delay sensitive	Non-delay sensitive	Non-delay sensitive
Connection	Connection-oriented	Connection-oriented	Connection-oriented	Connectionless
Utilization	Circuit	Variable bit	Connection-oriented	Connectionless
Emulation	Rate voice, video	Oriented data	Data
Relationship of ATM	Classes of Service	and AAL's

Economics

One of the possible questions in the mind of many readers is the economic trade-offs associated with the use of pure frame relay versus frame relay over ATM. For many economic comparisons, you must examine the cost components of each service. However, this can be a bit difficult with certain comparisons because most frame relay operations do not include a cost component based on the quantity of data transmitted, which most ATM operators do. Another problem associated with comparing the use of the two technologies involves comparing frames and cells when changes are based on the quantity of data transmitted. To provide you with an indication of how you can compare apples and oranges by costing frame relay and ATM service usage, this author will use prices that Quest publicized in early 2000 for frame relay non-discard-eligible, ATM variable-bit-rate, real-time, and ATM available-bit-rate services. Those prices per Mbyte of traffic transmitted were $.04, $.03, $.012, and $.0055, respectively.

Frame Relay Cost
An 8-Kbps voice-digitization rate, we noted earlier in this book, is equivalent to 3.6 Mbytes/hour. If we assume voice is packetized in 70-byte segments, the 3.6 Mbytes equates to 51/429 frames (3.6 Mbytes/70 bytes per frame). Because each frame has 6 bytes of overhead, total overhead becomes 51/429 frames x 6 bytes/frame, or 308/574 bytes. Thus, an hour of 8Kbps digitized voice results in 3.6 Mbytes plus 308 Kbytes of overhead, for a total of 3.9 Mbytes. Based on Quest's rate of $.04 and $.03 per Mbyte for non-discard-eligible and discard-eligible transmission, the cost of an hour of digitized voice is either $.156 or $.117!

ATM cost Prior to computing the cost of ATM/ we need to determine the number of cells and cell overhead. Because we assumed voice was digitized into 70-byte frames, each frame must be converted into two 53-byte cells. Thus, the 51/429 frames become 102/858 cells. Since each cell is 53 bytes/ the total amount of data/ including cell overhead, pay-load, and cell pads, becomes 102/858 cells x 53 bytes/cell, or 5.45 Mbytes. Because we assumed the per-Mbyte cost of ATM to be $.012 for VBR-rt and $.0055 cents for ABR/ the cost per hour for each type of ATM class of traffic becomes 5.45 Mbytes x $.012/Mbyte, or $.0654, and 5.45 Mbytes x $.0055/Mbyte, or $.03. Thus, in this example, if we do not consider the cost of equipment and focus our attention on transport charges, the use of an ATM backbone can result in significant savings.