ATM - Concepts and Architecture

Cell-Relay provides a compromise between fixed synchronous allocation mechanisms and bursty, routable packet interfaces.

The Asynchronous Transfer Mode (ATM) protocols and architecture have managed to gather an impressive amount of market and media attention over the last several years. Intended as a technique to achieve a working compromise between the rigidity of the telecommunication synchronous architecture and packet network's unpredictable load behavior, ATM products are appearing for everything from high-speed switching to local area networking. ATM has caught the interest of both the telecommunications community as a broadband carrier for Integrated Services Digital Network (ISDN) networks as well as the computer industry, who view ATM as a strong candidate for high-speed Local Area Networking. This article covers the basic concepts involved in the ATM architecture.

At the core of the ATM architecture is a fixed length "cell." An ATM cell is a short, fixed length block of data that contains a short header with addressing information, followed by the upper layer traffic, or "payload." The cell structure, shown in Figure 1, is 53 octets long, with a 5 octet header, followed by 48 bytes of payload. While the short packet may seem to be somewhat inefficient in its ratio of overhead to actual data, it does have some distinct advantages over the alternatives. By fixing the length of each cell, the timing characteristics of the links and the corresponding network are regular and relatively easy to predict; predicting the dynamics of variable length packet switched networks isn't always easy. By using short cells, hardware based switching can be accomplished. Finally, the use of short cells provides an ability to transfer isochronous information with a short delay.

Figure 1 - ATM Cell Structure (UNI Format)

The information contained in the header of each cell is used to identify the circuit (in the context) of the local link, carries local flow control information, and includes error detection to prevent cells from being mis-routed. The remaining 48 octets are routed through the network to the destination using the circuit.

ATM has evolved over the last 5-10 years to include a wide range of support protocols. Routing and congestion management, have been particular areas of research. The early concepts of cell transfer networks revolved around the thought that users could "reserve" a pre-specified amount of traffic through a circuit on the network. Some amount of guaranteed throughput would be provided with an additional amount only as needed. Then, through this contract, traffic in excess of the pre-allocated bandwidth could be arbitrarily dropped if congestion problems occurred. However, the complexities of implementation have proven these techniques to be far too difficult. Several vendors have proposed flow control architectures that involve more active windowing protocols between the switches for data traffic.

ATM Architecture

As in the case of many large systems, there are a range of components and connections involved in the ATM networks. Figure 2 shows an example network architecture. All connections in the ATM network are point-to-point, with traffic being switched through the network by the switching nodes. Two types of networks are included in the ATM architecture, Public Networks and Private Networks. Private Networks, often referred to as Customer Premises Networks, are typically concerned with end-user connections, or bridging services to other types of networks including circuit switched services, frame relay, and voice subsystems. The interface between the components in the Private Networks is referred to as the Private User Network Interface (UNI). ATM also extends into the wider area Public Networks.

Interfaces between the Public and Private network switches conform to the Public UNI. Interfaces between the switches within the Public network are the Network Node Interface (NNI). Specifications for both the Public and Private UNI can be found in the ATM Forum's publication "ATM User-Network Interface (UNI) Specification." The private networks often permit the use of lower speed short haul interconnects that are useful in LAN environments, but not of great use in wider area public networks. Three types of NNI have been developed, NNI-ISSI that connects switches in the same Local Area Transport Area (LATA), the NNI-ICI, that connects ATM networks of different carriers (InterCarrier), and finally, a Private NNI that permits the connection of different switches in a private network.

Figure 2 - ATM Sample Network Architecture

Protocol Reference Model

There is more to the ATM standards than the ATM cell format alone. Specifications exist to describe acceptable physical signaling, call control, and upper layer payload formats. Figure 3 shows the hierarchy of protocols involved in ATM. Mapping roughly to layers 1 and 2 of the OSI model, ATM is broken into 3 distinct layers. At the bottom, several classes of physical layers have been adapted to support the different types of ATM applications. The ATM layer provides the cell-switching and routing services. Application services rely on the ATM Adaptation Layer (AAL) that serves two purposes, to provide a common framework for the segmentation and reassembly of larger data sets into the ATM cells and to provide service specific mechanisms for the transport of different types of data. Four different classes of traffic are supported by the AAL ranging from straight circuit switched data through packet mode applications. Many of the early implementations of ATM have been focused on the packet mode services, often as a backbone for Frame Relay services. Typically, the AAL should be viewed as an internal, software interface to bridge end-user services over ATM. There is typically a good bit of work required to bind other protocols to the ATM stack.

Figure 3 - ATM Protocol Architecture

Traffic Flow Through The Network

A two tiered addressing scheme is used with the following elements being involved in the addressing assignments:

Virtual Channel: A virtual channel represents the flow of a single network connection data flow between 2 ATM end users. The ATM standards define this as a unidirectional connection between 2 end-points on the network
Virtual Path: A virtual path is used to carry one or more virtual channels through the network. It is represented as a bundle of channels between the two end-points.

This logical grouping of paths and channels provides some flexibility in managing the addressing of the flow of information through an ATM network. Figure 4 shows a sample configuration of a network with an assortment of Virtual Paths and Virtual Channels. As can be seen in the figure, each virtual path contains one or more virtual channels. It is important to note that the actual numbers assigned to each of the paths and channels are used only to represent a Virtual Path or Channel segment that exists between two adjacent nodes of the network. These values are established when the actual Virtual Channel Connections (VCC) are established. The number of Paths and Channels over a single link are limited by the ATM cell format. This limitation helps to explain why there are differences between the UNI and NNI formats. The NNI format replaces the 4 bits of the Generic Flow Control indication with additional VPI bits, extending the number of possible paths over the NNI from 256 to 4096.

Figure 4 - Example ATM Circuit and Path Connections

Figure 5 shows the formats for the UNI and NNI cells. The fields in the ATM Cells are:

GFC - (Generic Flow Control) used only in the UNI format. No general purpose services have been assigned to this field, and it is significant for only the local site. This flow control information is not carried from end-to-end. Two modes have been used for GFC based flow control, "uncontrolled access" and "controlled access." in uncontrolled access, this field is set to all zeroes. In controlled access mode, this field is set when congestion has occurred. The receiving equipment will report instances in which the GFC has been set a significant number of times to Layer Management.
VPI/VCI - (Virtual Path Identifier/Virtual Channel Identifier). The distribution of bits between these two fields can be negotiated between the user and network equipment. Referring back to Figure 4, these identifiers are used to tag only the portion of the path/circuit connection over a single link. It is the combination of all of the individual paths and circuits that comprise the connection.
PT - (Payload Type) Indicates whether or not the cell contains use information or Layer Management Information. It also carries implicit congestion information.
CLP - (Cell Loss Priority). Indicates the cell's priority in the ATM selective loss algorithm. Set by the initiating equipment, when this set to 0, the cell is given preference over cells with CLP set to 1.
HEC - (Header Error Control). Provides a capability to correct all single bit errors in the cell header as well as the detection of the majority of multiple-bit errors. The use of this field is up to interpretation of the equipment designers. If most errors are likely to be single bit errors, it can be used for error correction. Using the field for error correction does carry some level of risk of introducing unwanted errant traffic on the network should a mistake be made in the correction process.

Note that the circuit and path identification fields are used to indicate the path that each cell is to take through the network. The identifiers carried in the cells carry only the information required to identify the cell's route to the receiving switch or end-point, they are not network addresses as is found in the case of IP or OSI networks.

Figure 5 - ATM Cell Formats

This article covers some of the important general concepts in the ATM architecture, but scratches just the surface. Other important areas of the ATM architecture include how it is mapped to the various physical interfaces, the ATM Adaptation Layer, signaling protocols, layer management, along with switching strategies.