2.4 Equal-Sized Packets Model: ATM
A networking model in which packets are of equal size can be constructed. Equalsized packets, or cells, bring a tremendous amount of simplicity in the networking hardware, since buffering, multiplexing, and switching of cells become extremely simple. However, a disadvantage of this kind of networking is the typically high overall ratio of header to data. This issue normally arises when the message size is large and the standard size of packets is small. As discussed in Section 1.3, the dominance of headers in a network can cause delay and congestion. Here, we describe Asynchronous Transfer Mode technology as an example of this model.
The objective of Asynchronous Transfer Mode (ATM) technology is to provide a homogeneous backbone network in which all types of traffic are transported with the same small fixed-sized cells. One of the key advantages of ATM systems is flexible multiplexing to support multiple forms of data. ATM typically supports such bursty sources as FAX, coded video, and bulk data. Regardless of traffic types and the speed of sources, the traffic is converted into 53-byte ATM cells. Each cell has a 48-byte data payload and a 5-byte header. The header identifies the virtual channel to which the cell belongs.
Similar to a telephone network, ATM is a set of connection-oriented protocols, which means that a connection must be preestablished between two systems in a network before any data can be transmitted. ATM is capable of supporting and integrating data, voice, and video over one transmission medium with high bit data rate delivery services into a single network. ATM is bandwidth-on-demand networking and is scalable in bandwidth with the ability to support real multimedia applications.
The use of fixed-size cells can greatly reduce the overhead of processing ATM cells at the buffering and switching stages and hence increase the speed of routing, switching, and multiplexing functions. However, the great ratio of header to data makes the technology unsuitable for wide area networks and limits its applications in small networks. Like IPv6, ATM supports QoS mainly to reserve resources that guarantee specified maximum delay, minimum throughput, and maximum data loss. The QoS support allows ATM to concurrently handle all kinds of traffic.
ATM connections are identified by a virtual channel identifier (VCI) and a virtual path identifier (VPI). VCI and VPI are combined to be used in a switch to route a cell. As shown in Figure 2.11, the identity of a "physical" link is identified by two "logical" links: virtual channel (VC) and virtual path (VP). When a connection is set up, the values of these identifiers remain unchanged for the lifetime of the ATM connection.
Figure 2.11 Overview of a typical ATM transmission medium
Example. Figure 2.12 shows a routing table in an ATM switch, with routing information for all active connections passing through the switch. The routing information consists of the new VPI/VCI and new outgoing link for every incoming VC. Link 5, with VPIs 1, 3, and 5, can be switched on link 10 with VPIs 3, 7, and 8 through the ATM switch. A routing table provides the detail of the switching function. For example, a cell with VPI 3 and VCI 9 on link 5 is set to be forwarded with VPI 7 and VCI 2 on link 10.
Figure 2.12 A routing table in an ATM switch
2.4.1 ATM Protocol Structure
The ATM protocol structure is shown in Figure 2.13. The three-dimensional model includes four layers in the vertical dimension. The tightly linked layers consist of the physical layer, the ATM layer, the ATM adaptation layer (AAL), and higher layers. The physical layer includes two sublayers: the physical medium and transmission convergence. The physical medium sublayer defines the physical and electrical/optical interfaces with the transmission media on both the transmitter and the receiver. This layer also provides timing information and line coding. The transmission convergence sublayer provides frame adaptation and frame generation/recovery.
Figure 2.13 ATM protocol reference model
The ATM layer provides services, including cell multiplexing and demultiplexing, generic flow control, header cell check generation and extraction, and most important, remapping of VPIs and VCIs. The AAL layer maps higher-layer service data units, which are fragmented into fixed-size cells to be delivered over the ATM interface. In addition, this layer collects and reassembles ATM cells into service data units for transporting to higher layers. The four types of AALs support different classes of services.
- AAL1 supports class A traffic, the required timing between a transmitter and a receiver, and the constant bit rate (CBR) traffic.
- AAL2 supports class B traffic and time-sensitive—between source and sink—but variable bit rate (VBR) data traffic.
- AAL3/4 supports class C or class D traffic and VBR data traffic.
- AAL5 supports class D traffic in which VBR traffic can be transported and no timing relationship between source and sink is required.
The higher layers incorporate some of the functionality of layers 3 through 5 of the TCP/IP model. The control plane at the top of the cube shown in Figure 2.13 involves all kinds of network signaling and control. The user plane involves the transfer of user information, such as the flow-control and error-control mechanisms. The management plane provides management function and an information-exchange function between the user plane and the control plane. The management plane includes (1) plane management that performs management and coordination functions related to a system as a whole, and (2) the layer management that monitors bit error rates on a physical communications medium.
An ATM network can support both a user-network interface (UNI) and a network-node interface (NNI). A UNI is an interface connection between a terminal and an ATM switch, whereas an NNI connects two ATM switches. A summary of these two interfaces is shown in Figure 2.14. If privately owned switches are in the network, the interface between the public and private parts of the network is called the public NNI, and the connection between two private switches is known as the private NNI (P-NNI).
Figure 2.14 Overview of signaling: (a) UNI format; (b) NNI format
2.4.2 ATM Cell Structure
An ATM cell has two parts: a 48-byte payload and a 5-byte header, as shown in Figure 2.15. The choice of a 48-byte payload was a compromise among various design teams considering two important factors: packetization delay and transmission efficiency. (Refer back to Section 1.3 on packet size optimization.) The header consists of several fields. However, the ATM cell header has two different formats: UNI and NNI. The details of the UNI 5-byte header are as follows.
- The 4-bit generic flow control (GFC) field is used in UNI only for controlling local flow control. This field enables the participating equipment to regulate the flow of traffic for different grades of service. Two modes are defined for this field: the controlled GFC, to provide flow control between a user and a network, and the uncontrolled GFC, to indicate that the GFC function is not used.
- Together, the virtual path identifier and the virtual channel identifier represent an ATM address. A VPI identifies a group of virtual channels with the same end point. A VCI identifies a virtual channel within a virtual path.
- The 3-bit payload type field is used to indicate the type of data located in the payload portion of the cell: for example, 000, the current cell is a data cell and no congestion is reported; 010, this is a user data cell, and congestion is experienced. The payload could be congestion information, network management message, signaling information, or other forms of data.
- The 1-bit cell-loss priority (CLP) field is used to prioritize cells. When congestion occurs, cells with CLP set to 1 (considered low priority) are discarded first. If the bit is set to 0, the cell gets higher priority and should be discarded only if it could not be delivered.
- The 8-bit header error control (HEC) field is used for error checking. HEC functions include correcting single-bit errors and detecting multiple-bit errors.
Figure 2.15 An ATM cell and its header structure
The main difference between NNI and UNI formats is that the 4 bits used for the GFC field in the UNI cell header are added to the VPI field in the NNI cell header. Thus, for NNI, VPI is 12 bits, allowing for more VPs to be supported within the network. VCI is 16 bits for both cases of UNI and NNI. The values of VPI and VCI have local significance only with a transmission link. Each switching node maps an incoming VPI/VCI to an outgoing VPI/VCI, based on the connection setup or routing table, as shown in Figure 2.12.
HEC works like other checking methods. First, the transmitter calculates the HEC field value, and the receiver side runs an algorithm consisting of two modes of operation. At initialization step, this field starts with an error-correction mode. If a single-bit error is detected in the header, the error-correction algorithm identifies the error bit and then corrects it. If a multibit error is detected, the mode moves to detection mode; errors are discarded but not corrected. Error-detection mode remains whenever cells are received in error, moving back to correction mode only when cells are received without error.