2. ATM

Asynchronous Transfer Mode (ATM) is a low-level communications protocol designed with the intention of supporting a wide variety of applications including voice, video, and data communications. ATM is based on the Broadband Integrated Services Digital Network (B-ISDN). While originally defined by the International Telecommunications Union Telecommunications standards body (ITU-T), recent progress of ATM standards has been pushed by the ATM Forum in the United States. The ATM Forum is a consortium of over 750 companies as well as government agencies and research organizations.

2.1 ATM Overview

ATM may be formally defined as an "interface and protocol designed to switch variable bit-rate and constant bit-rate traffic over a common transmission medium"[1]. ATM supports both variable and constant bit rate traffic. Four ATM adaptation layers are defined to support different traffic types and different classes of service. Different service classes offer trade-offs between parameters such as latency and error requirements. Service classes are discussed in more depth later in this section.

While ATM supports both connection-oriented and connection-less services, ATM is itself a connection-oriented protocol. Data transmission is handled by establishing Virtual Paths (VP) and Virtual Channels (VC) over a transmission path. A transmission path contains one or more virtual paths, and each virtual path contains one or more virtual channels. ATM switches can switch traffic at both the VC level and the VP level (i.e. switch multiple VCs at once).

2.1.1 ATM Protocol Layers

Figure ATM-1 shows the ATM protocol layers. The ATM Physical layer maps to the Open Systems Interconnect Reference Model (OSI/RM) physical layer, while the ATM and AAL layers correspond (loosely) to the OSI/RM data link layer.(note 1) [2]

The Clear Channel Transmission Convergence specification "provides a means to transport ATM cells in the payload of any available transport mechanism. Examples include V.35, EIA/TIA 449/530, EIA/TIA 612/613 (HSSI) or even RS-232[c]." [3]

The TC layer is responsible for cell rate de-coupling, cell delineation for synchronous interfaces, as well as transmission frame generation/recovery. Cell rate de-coupling is required whenever a synchronous interface, such as a DS3 (T3) carrier, is used. Idle cells may be inserted into the transmission stream if the source does not provide data as fast as the carrier. Cell delineation allows the receiver to know where a cell starts and ends in the continuous bit stream.

2.1.1.2 ATM Layer

The ATM layer provides for VP/VC translation (i.e. switching), cell multiplexing, as well as cell-generation. Each of these functions are described in the following paragraphs.

ATM is a connection-oriented protocol, i.e. a point-to-point connection is setup between a transmitter and a receiver before user data is transferred. Each connection is identified by a Virtual Path Identifier (VPI) and a Virtual Channel Identifier (VCI). Each VPI/VCI pair is unique only between the transmitter and the receiver, so that an end-to-end connection between workstations may have several VPI/VCI pairs. An ATM switch will translate data on an incoming VPI/VCI to an outgoing VPI/VCI. Switching may occur at the VPI level (only the VPI is translated), thus all VCIs in a VP are switched at once. Alternatively, switching may occur at the VCI level, in which case both the VPI and VCI are translated.

2.1.1.3 AAL Layer

To support the different service classes, the AAL layer is divided into two sublayers, Segmentation and Re-assembly (SAR) and the Convergence Sublayer (CS). The CS is further divided into Service Specific (SSCS) and Common Part (CPCS) sublayers. The SAR is responsible for segmenting the user data into 48-byte protocol data units (PDU), which are then placed in the payload field of each cell by the ATM layer. The format of the PDU varies, depending on the service class employed.

The CPCS is required for all services, and defines the common PDU format for a given service. The CPCS layer will take the SSCS PDU (which is simply the user data if SSCS is null) and places it in the CPCS payload field, adding headers and/or trailers as needed for the type of service.

2.1.2 ATM Service Classes

Table ATM-2 shows the mapping of services classes to AAL layers. The reader will note that Table ATM-2 gives two names for each class, the former is the ITU-T class name while the latter is the equivalent ATM Forum class name.

ATM Service Classes

As mentioned previously, the different service classes enable ATM to support a wide variety of applications. Class A is designed for voice (telephone) services which require a constant bit-rate (CBR) and low latency (time-sensitive). Class B is intended for applications such as video teleconferencing, which also require low latency, but have variable data rates (VBR). Data communications, such as file transfers or transactions, do not have strict time requirements, but cannot tolerate errors (loss-sensitive). Class C is meant for connection-oriented data services, such as Frame Relay, while Class D is intended for connectionless data such as IP or Switched Multi-megabit Data Service (SMDS) traffic. Class C is also referred to as variable bit rate, non-real time (VBR-NRT) to distinguish it from the real time, variable bit rate service of Class B (VBR-RT).

The latter two classes are differentiated from the first three by Quality of Service (QoS) guarantees (note 2). For Unspecified Bit Rate (UBR) data, such as legacy LAN traffic, the ATM network does not provide any QoS guarantees or traffic management. Available bit rate (ABR) is a "best effort" service, in which the network makes use of flow control mechanisms to increase the bandwidth available to the user. When there is congestion, the available bandwidth is reduced.

2.2 ATM as a Wide Area Network

ATM has its roots in wide-area networking, specifically the Broadband-ISDN standards developed by the ITU-T. As covered in section 2.1, ATM supports a variety of interfaces, many of which are WAN-based. Indeed, one of the many benefits touted by ATM enthusiasts is the seamless integration of LANs and WANs since ATM can be used in both areas. For wide area networking, there are two sets of telecommunication standards for which ATM interfaces have been designed for, namely the Plesiochronous Digital Hierarchy (PDH) and the Synchronous Digital Hierarchy (SDH). The PDH standard defines the rates upon which the T-carriers are based. The SDH standard defines the rates which Synchronous Optical NETwork (SONET) carriers are based (OC-1, OC-3, etc.).

The ATM Forum has defined several specifications for interfacing ATM over the PDH and SDH carriers, including the Data eXchange Interface (DXI), the Frame relay User Network Interface (FUNI), and the Broadband InterCarrier Interface (B-ICI). The DXI and FUNI are ATM access protocols run ATM over low cost frame-based user equipment. The B-ICI specification defines a standard interface so that different service providers can connect to each other's networks. The ATM DXI specification supports the V.35, RS-449, and High Speed Serial Interface (HSSI) physical layer standards at speeds ranging from a few Kbps up to 50 Mbps.

2.3 ATM over Satellite

Much research has been done analyzing the performance of ATM over satellite links [4,5,6]. In particular, NASA Lewis Research Center has conducted numerous experiments over the Advanced Communications Technology Satellite (ACTS) satellite [7,8,9,10]. The ACTS satellite network has been linked to several terrestrial ATM networks, including ATDnet, MAGIC, AAI, and AAMnet [11,12,13,14]. Each experimental network supports on-going research on various topics related to ATM.

Satellite communications pose several issues for ATM, namely the long propagation delays and the higher bit error rates. ATM was designed for fiber optic transmission lines, which have low propagation delays, and BER on the order of 10^-10. More importantly, errors in fiber occur randomly, rather than in bursts. Therefore, minimal error correcting mechanisms are defined in the ATM protocol (note 3). However, in a satellite environment, the error characteristics are such that errors occur in bursts and affect the payload as well as the header. One solution to this problem has been developed and is commercially available. The ATM Link Enhancer (ALE) from Comsat interleaves the ATM cells such that a burst error typically only causes single bit errors in the interleaved cells [15].

The long propagation delay primarily impacts the higher layer protocols, such as Transmission Control Protocol (TCP). The performance of TCP over satellite may be improved by implementing the recommendations published in the Internet Engineering Task Force (IETF) Request For Comments (RFC) documents. Specifically, RFC 1323 (large windows), RFC 2001 (congestion avoidance), and RFC 2018 (selective acknowledgment) have been shown to greatly improve TCP performance over satellite links [16].

Other systems may not use delay-sensitive protocols, such as raw data collection from scientific instruments on board a satellite. In such cases, the data communications may be implemented using ATM onboard and SONET over the satellite link. Only the bit error rates need be addressed, since the propagation delay is no longer a factor.

To date, experiments of ATM over satellite have only used the satellite as a "bent-pipe" for data communications. The more interesting case, and one in which the industry is moving, views the satellite as a network node with on-board switching capabilities. This is the goal of several modern satellite networks, which seek to provide ubiquitous voice and data communication services [17,18,19]. Further, Kim et.al. describe work in developing an on-board ATM switch, using the IEEE 1355 standard [20].

NOTES:

1. This is a matter of some debate. Several aspects of ATM are more characteristic of the network layer (hierarchical address space, routing protocol). Therefore an exact correspondence between the ATM model and the OSI/RM is not possible.

2. QoS is a method to categorize serivces based on bandwidth, delay (latency) and loss requirement. Traffic management provides for control of data transfers (e.g. flow control, congestion control).

3. The Header Error Control (HEC) is an 8-bit field in the ATM cell header that provides single bit error correction over the header. The error checking does not cover the payload. ATM leaves the responsibility for error recovery on the payload to higher level protocols (e.g. TCP).