Asynchronous-transfer mode (ATM) is a core telecommunications technology that evolved from analog voice service, X.25 and frame relay. ATM is the evolutionary technology that is best able to fulfill the quality-of-service (QoS) requirements for today's isochronous applications. It is likely that IP will eventually be able to make similar claims, but there are still issues to be resolved before streaming broadcast video can be easily carried across public IP networks.
 Figure 1. This comparison between TDM and PSM shows how ATM uses PSM to service high-priority traffic first. Image courtesy SBC-TRI. |
Voice networks started as analog transmissions over copper wires using tube-type line amplifiers and mechanical crossbar switches. Such early voice networks had several problems. They were error-prone, expensive to lease and relatively unreliable for transmitting data between facilities. But, at the time, it was the best technology available. With the advent of solid-state electronics and integrated circuits, this service was converted to digital. The use of pulse-code modulation (PCM) permitted the multiplexing of many voice lines onto a single circuit.
X.25 networks were developed to address the shortcomings of analog networks. X.25 is a layer-3 protocol that guarantees data integrity through store and forward, error checking, and hop-by-hop acknowledgment. X.25 was the first network technology that allowed customers to pay only for the traffic they generate.
Frame-relay networks are based upon layer-2 protocols, and were originally designed to achieve speeds up to 45Mb/s. Frame-relay switches work in round-robin fashion, so they can't service traffic by priority and type. Frame relay also does not provide QoS parameters that establish service guarantees. Finally, frame relay creates switch overhead because frame boundaries are not predictable.
ATM, however, differs significantly from these predecessors. It allocates bandwidth fairly and prioritizes traffic based on the needs of the application. A full range of QoS parameters are available to create a Service Level Agreement (SLA) or contract for service across a connection. This is extremely important for any public network. Also, ATM has small fixed-length cells that switch faster with deterministic behavior. Some may argue that ATM's small cell size is inappropriate for video. They say small cells lead to inefficiency because the ratio of header information to payload is relatively high. This argument is correct in local network installations. But the flexibility and speed afforded by the smaller cell size across a nationwide public switched network greatly outweigh the cost of the header overhead.
As its name implies, ATM permits asynchronous networking. This contrasts with networks that use time-domain multiplexing (TDM). Figure 1 shows that TDM allocates a time slot to each piece of multiplexed traffic on a periodic (synchronous) basis, whereas the prioritized statistical multiplexing (PSM) used in ATM gives higher-priority traffic more time slots based on a well-defined algorithm.
ATM is connection-oriented, although it also provides connectionless protocols outside the network. When operated as a connection-oriented protocol, ATM requires a virtual connection (either a permanent virtual circuit or a switched virtual circuit) before the network can accept traffic. During the ATM setup time, switches allocate bandwidth to the virtual circuit for the duration of the “call.” This is another key aspect of ATM for broadcast. An ATM virtual circuit configured with the proper QoS parameters has a guaranteed bandwidth available to handle the video connection.
ATM circuits can handle a mix of traffic types, including voice, data and video. The user requirements for each type are different. With voice traffic, the bandwidth requirement is low (less than 64kb/s) and users can talk at any time. But there is very little tolerance for delay during transmission. Studies have shown that delays over about 250 milliseconds severely impact the ability of people to talk naturally over a voice link.
Data transmissions over ATM can be based on connection or on file transfer. Unlike broadcast video, ATM buffers computer data and transmits it in bursts when bandwidth becomes available. Bandwidth requirements vary widely depending on the customer and application. Compared to voice or video, data applications generally are much more tolerant to delay, jitter and other network disturbances.
 Figure 2. ATM is a layer-2 protocol that specifies the data link and physical layers in the OSI model.Image courtesy SBC-TRI. |
Video traffic, especially video for professional applications, has some tight constraints. The transmission is continuous. Thus, at the output of the network-edge device, data must appear in the proper order and meet video's requirements for delay, jitter and wander. Real-time interactive video requires low delay and tighter error control. Furthermore, if the video is compressed and transmitted at a variable bit rate (VBR), the bandwidth required for video transmission can be highly variable and bursty.
In the ATM world, QoS can specify various service parameters. QoS allows the network to support the service needs of different types of traffic. It can accommodate multiple traffic types with differing user requirements over a single infrastructure. It services and prioritizes traffic in the network on a per-virtual-connection basis. There are several QoS service classes, grouped roughly according to user requirements. These include constant bit rate (CBR), variable bit rate — real time (VBR-rt), variable bit rate — non-real time (VBR-nrt), available bit rate (ABR) and unspecified bit rate (UBR).
ATM is a layer-2 protocol. Figure 2 shows how ATM specifies the data link and physical layers in the OSI model. The ATM adaptation layer (AAL) maps traffic into the cell payload. The ATM layer adds control data in the cell header. The physical layer maps cells to transmission media, typically synchronous optical network (SONET) on switched public networks. The customer-premises equipment (CPE) takes the network output of a device and, using AAL, maps the data onto the ATM network. Once inside the network, the ATM switches establish connections and route cells using the information in the ATM cell headers. The information is then extracted from the ATM cells at the other end using ATM customer-premises equipment and presented to the network interface of the remote application.
ATM cells have a 48-byte payload. To get the user data broken down into these payloads, the AAL contains a convergence sublayer (CS), which provides application-specific timing and control. The segmentation-and-reassembly (SAR) rules break down CS protocol data units into 48-byte cell payloads (see Figure 3).
 Figure 3. The AAL layer distributes application data into 48-byte cell payload blocks. |
There are several standardized ATM adaptation layers. AAL-1 is for real-time, constant bit rate applications. It is primarily used for circuit emulation. Each cell contains an AAL-1 header. For example, AAL-1 can be used to map SDI into ATM. AAL-2 is for real-time variable-bit-rate applications. It is not widely used because most developers have found AAL-5 better suited and less complicated to use. AAL-3 and AAL-4 (connection-oriented and connectionless, respectively) have been merged into AAL-3/4 and are primarily used in Europe. They have not been widely adopted elsewhere due to high payload overhead and packet-interleaving complexity. AAL-5 is used for most traffic types other than CBR. It has minimal control-field overhead, and it automatically pads cells to ensure 48 bytes per cell.
The ATM layer appends a four-byte header for address and control purposes. There are two types of ATM cell headers: user-to-network interfaces (UNI) and network-to-network interfaces (NNI). The ATM layer multiplexes cells from various applications (voice, data or video) into a single cell stream for presentation to the physical layer.
The physical layer is comprised of two sublayers: the transmission-convergence sublayer (TCS) and the physical-medium-dependent (PMD) sublayer. The TCS is responsible for controlling header errors, mapping cells into transport payloads and delineating cell boundaries. The PMD is the specification of the physical connections required by the specific physical medium.
ATM also supports traffic management (TM), which allocates network resources fairly and predictably for individual connections as specified by their QoS parameters. TM is enforced using traffic policing and shaping, connection-admission control (CAC), and congestion management. TM is one of the key capabilities that allows ATM to carry broadcast video over public networks. Without it, it would be impossible to guarantee that video traffic would be delivered at a consistent quality across the network, regardless of other traffic loads.
Three key QoS parameters for video service are cell-delay variation (CDV), cell-transfer delay (CTD) and cell-loss ratio (CLR). CDV is the variation in time between when you expect a cell to arrive and when it actually shows up. CTD is the overall delay across an ATM network. This is deterministic and predictable. CLR is the lost cells divided by the total transmitted cells. There is ongoing work in the Video Services Forum to determine the impact of CDV, CTD and CLR on video delivered across ATM networks.
As you learn about ATM, you may feel overwhelmed by three-letter acronyms (TLAs). Fortunately, there is an excellent glossary published on the Internet by the ATM Forum. You can find it at www.dit.upm.es/snh/arhelp/glossaries/atmf/.
Brad Gilmer is president of Gilmer & Associates, executive director of the AAF Association, and executive director of the Video Services Forum.
Professional, high-resolution, high-bandwidth video
Intolerant of cell delay and cell loss
Video conferencing and lower-quality professional video
Medium bandwidth, more tolerant of cell delay
Videotape feeds
Higher bandwidth, no cell-delay concern
Some cell-loss toleration because of retransmit capabilities of end devices
Computer data, videophone feeds
No cell-delay concern, some cell-loss toleration because of retransmit capabilities of end devices
Send questions and comments to: brad_gilmer@primediabusiness.com