10.3 Asynchronous Transfer Mode

---

ATM is a cell-switching technology that offers low-latency transmission with built-in QoS guarantees in support of data, voice, and video traffic at multi-mega-bit-per-second speeds. The high speed is the result of ATM’s fixed-length, 53-byte cell, which is switched in hardware. ATM is also highly scalable, making it equally suited for interconnecting legacy systems and LANs and for building WANs over today’s high-performance optical-fiber infrastructures. ATM-based networks may be accessed through a variety of standard T-carrier and OC interfaces, as well as such services as frame relay and IP.

10.3.1 Applications

Many applications are particularly well suited for ATM networks, including the following:

• LAN internetworking: ATM can be used to interconnect LANs over the WAN. Special protocols for LAN emulation (LANE) make the connection-oriented ATM network appear as a connectionless Ethernet or token-ring LAN segment.

• Videoconferencing or broadcasting: ATM can be provisioned for interactive video-conferencing between two or more locations or to support point-to-multipoint video broadcasts.

• Telemedicine: With ATM, large amounts of bandwidth can be provisioned to support the rapid exchange of high-resolution diagnostic images and multimedia patient records while permitting interactive consultations among medical specialists at different locations.

• Private-line connectivity: An ATM virtual circuit can be used to provide a more economical way to provision leased lines on the WAN. ATM protocols can emulate N x 64 Kbps DS0 transport.

• PBX voice trunking: An ATM virtual circuit can be used to interconnect PBXs and maintain full PBX feature support, call routing, and switching. Voice trunking combines multiple calls onto a single virtual circuit for further bandwidth optimization, reduced delay, and lower cost. PBX voice trunking requires an integrated access device at the customer premises, between the PBX and ATM switch, which performs the protocol conversions necessary to extend feature signaling across the ATM network.

ATM also offers a consolidation solution for any company that maintains separate-networks for voice, video, and data. The reason for separate networks is to provide appropriate bandwidth and preserve performance standards for the different applications. But ATM can eliminate the need for separate networks, providing a unified platform for multiservice networking that meets the bandwidth and QoS needs of all applications. Although the start-up cost for ATM is high, the economics of network consolidation mean that companies do not have to wait very long to realize significant return on their investment.

10.3.2 Quality of Service

ATM serves a broad range of applications very efficiently by allowing an appropriate QoS to be specified for each application. Various categories have been developed to help characterize network traffic, each of which has its own QoS requirements. These categories and QoS requirements are summarized in Table 10.1.

Table 10.1: ATM Quality of Service Categories

		Quality of Service
Category	Application	Bandwidth Guarantee	Delay Variation Guarantee	Throughput Guarantee	Congestion Feedback
Constant bit rate (CBR)	Provides a fixed virtual circuit for applications that require a steady supply of bandwidth, such as voice, video, and multimedia traffic.	Yes	Yes	Yes	No
Variable bit rate (VBR)	Provides enough bandwidth for bursty traffic such as transaction processing and LAN interconnection, as long as rates do not exceed a specified average.	Yes	Yes	Yes	No
Available bit rate (ABR)	Makes use of available bandwidth and minimizes data loss through congestion notification. Applications include e-mail and file transfers.	Yes	No	Yes	Yes
Unspecified bit rate (UBR)	Makes use of any available bandwidth for nonessential communications between computers, such as e-mail delivery acknowledgements, but does not guarantee when or if data will arrive at its destination.	No	No	No	No

CBR is intended for applications where the PVC requires special network timing requirements (i.e., strict PVC cell loss, cell delay, and cell-delay variation performance). For example, CBR would be used for applications requiring circuit emulation (i.e., a continuously operating logical channel) at transmission speeds comparable to DS1 and DS3. Such applications would include private-line-like service or voicetype service where delays in transmission cannot be tolerated.

Variable bit rate-real time (VBR-rt) is intended for applications where the PVC requires low cell-delay variation. For example, VBR-rt would be used for applications such as VBR video compression and packet voice and video, which are somewhat tolerant of delay. Variable bit rate-non real time (VBR-nrt) is intended for applications where the PVC can tolerate larger cell delay variations than VBR-rt. For example, VBR-nrt would be used for applications such as large file transfers.

ABR is intended for routine applications and when the customer seeks a low-cost method of transporting bursty data for noncritical applications that can tolerate delay variations. The traffic goes out into the network when bandwidth becomes available; otherwise, it is held back until other applications with higher priority are finished using the bandwidth. If congestion builds up in the network, ABR traffic may be held back at the CPE to help relieve the congestion condition.

UBR is also intended for routine applications and when the customer seeks a low-cost method of transporting bursty data for noncritical applications that can tolerate delay variations. Although the carrier will attempt to deliver all ATM cells received over the PVC, if there is any network congestion, this traffic is the first to be discarded to relieve the congestion.

All of these QoS parameters can be set up by the corporate network manager using the Windows-based GUI that comes with the router or integrated access device. If the ATM service is managed by the carrier, then the carrier will supply the CPE, program it per the performance requirements of the customer’s applications, install it, and maintain it.

QoS enables ATM to admit a CBR voice connection, while protecting a VBR connection for a transaction-processing application, and allowing an ABR or UBR data transfer to proceed over the same network. Each virtual circuit will have its own QoS contract, which is established at the time of connection setup at the user-to-network interface (UNI). The network will not allow any new QoS contracts to be established if they will adversely affect its ability to meet existing contracts. In such cases, the application will not be able to get on the network until the network is fully capable of meeting the new contract.

As noted, when the QoS is negotiated with the network, there are performance guarantees that go along with it: maximum cell rate, available cell rate, cell transfer delay, and cell loss ratio. The network reserves the resources needed to meet the performance guarantees, and the customer is required to honor the contract by not exceeding the negotiated parameters. Several methods, including traffic policing and traffic shaping, are available to enforce the contract. Traffic policing is a management function performed by switches or routers on the ATM network. To police traffic, the switches or routers use a buffering technique referred to as a “leaky bucket.” This technique entails traffic flowing (leaking) out of the buffer (bucket) at a constant rate (the negotiated rate), regardless of how fast it flows into the buffer. If the traffic flows into the buffer too fast, the cells will be allowed onto the network only if enough capacity is available. If there is not enough capacity, the cells are discarded and must be retransmitted by the sending device.

Traffic shaping is a management function performed at the UNI of the ATM network. It ensures that traffic matches the contract negotiated between the user and the network during connection setup. Traffic shaping helps guard against cell loss in the network. If too many cells are sent at once, cell discards can result, which will disrupt time-sensitive applications. Because traffic shaping regulates the data transfer rate by evenly spacing the cells, discards are prevented.

10.3.3 Virtual Circuits

As in frame relay, ATM virtual circuits can be bidirectional or unidirectional, meaning that each virtual circuit can be configured for one-way or two-way operation. The virtual circuits can be configured as point-to-point (i.e., PVC), switched, or multipoint. They can also be symmetric or asymmetric in nature. In other words, each bidirectional virtual circuit can be configured for symmetric operation (same speed in both directions) or asymmetric operation (different speeds in each direction).

A virtual circuit has two components: a virtual path and a virtual channel. In this simplified view of an ATM network (see Figure 10.1), the customer has two locations connected together by a virtual path, which contains a bundle of virtual channels. Each of the three virtual channels is assigned to a particular end system, such as a PBX, server, or router.

Figure 10.1: A simplified view of virtual circuits through an ATM network.

In this example, an IAD is used to consolidate the virtual circuits and deliver them to the ATM switch via a dedicated line. VC-1 provides LAN users with access to a mainframe. VC-2 provides trunking between two PBXs. VC-3 provides LAN users with access to a remote server.

In a large network, there may be hundreds of virtual paths. ATM standards allow up to 65,000 virtual channels to share the same virtual path. This scheme simplifies network management and network recovery. When a virtual path must be reconfigured to bypass a failed port on an ATM switch, for example, all its associated virtual connections go with it, eliminating the need to reconfigure each virtual circuit individually.

On a private ATM network, the individual end devices at each location can be identified by IP addresses or, in the case of LAN emulation, MAC addresses. For interoperability between private and public networks, several addressing formats are in use. One of them is based on the E.164 standard, which entails giving each device its own 15-digit address, which is similar to a telephone number used on the PSTN. (Internationally, the ITU Telecommunication Standard Board assigns country/service codes.) Blocks of E.164 addresses are obtained from the numbering plan administrator in each country for assignment by local service providers to their subscribers.

To facilitate network administration, the ATM Forum’s Interim Link Management Interface (ILMI) automates address registration. Any device joining the network exchanges address information with its host switch to determine the complete ATM address. If the same E.164 number is used for the same service from multiple providers, other criteria need to be used for determining which service provider’s database will be consulted for call delivery.

10.3.4 ATM Layers

Like other technologies, ATM uses a layered protocol model. ATM exists at Layer 2 of the OSI model and has only four layers (see Figure 10.2), which typically operate above SONET at Layer 1.

Figure 10.2: ATM protocol model in relation to the OSI reference model and SONET physical layer protocols.

The ATM adaptation layer (AAL) provides the necessary services to support the higher-layer protocols. The functions of this layer are summarized in Table 10.2 and may exist in the end-stations, servers, or network switches. Among other things, this layer is responsible for segmenting the information into 53-byte cells and, at the receiving end, reassembling it back into its native format. This is known as segmentation and reassembly (SAR). Table 10.2 describes the adaptation layers.

Table 10.2: ATM Adaptation Layer Functions AAL Application Examples Notes

AAL	Application Examples	Notes
1	Isochronous constant bit-rate services, such as circuit emulation.	Supports connection-oriented services that require CBRs and have specific timing and delay requirements.
2	Mobile trunking, which entails the transmission of compressed voice between radio base stations and a mobile switching center.	Supports low-bit-rate delay-sensitive traffic
3/4	Near real-time VBR data, either connectionless or connection-oriented.	Originally intended as two layers: one for connection-oriented services and the other for connectionless services. AAL 3/4 is no longer used; AAL5 satisfies all potential applications.
5	Near real-time VBR data, such as datagrams and signaling messages.	Supports connection-oriented VBR data services. It is a leaner AAL compared with AAL 3/4 at the expense of error recovery and built-in retransmission. It uses less bandwidth overhead, has simpler processing requirements, and reduces implementation complexity.

10.3.5 Inverse Multiplexing over ATM

Today, even mid-size companies with multiple traffic types and three or more distributed locations can benefit from ATM’s sustained throughput, low latency, and adept traffic handling via appropriate QoS mechanisms. The availability of ATM-based inverse multiplexers and N x T1 access makes ATM suitable for mainstream use, particularly for companies that appreciate the benefits of ATM but have been locked out of the service due to its high cost of implementation.

In the past, T3 links were the minimum bandwidth required to access ATM networks, making the cost prohibitive for the vast majority of companies. Inverse multiplexing over ATM (IMA) solves the bandwidth gap problem. With IMA, companies can aggregate multiple DS1 circuits to achieve just the right amount of bandwidth they need for their applications and pay for only that amount on an N x T1 basis. The advantage of IMA is that such companies can scale up to the bandwidth they need, starting with a single T1, and then add links as more bandwidth is justified.

For example, when the bandwidth of four T1s is bonded by the IMA device, the virtual connection through the service provider’s network is provisioned at 6 Mbps. When the bandwidth of eight T1s is bonded by the IMA device, the virtual connection through the service provider’s network is provisioned at 12 Mbps. Regardless of the number of T1 access links in place, the IMA device bonds them together, combining the bandwidth into a fatter logical pipe that can support mixed-media applications running over interconnected LANs (see Figure 10.3).

Figure 10.3: IMA allows the use of bonded multiple T1 access lines into and out of the service provider’s ATM network, rather than forcing companies to use more expensive T3 access lines at each location. This makes it cost-effective for mid-size companies to take advantage of ATM services to support a variety of applications.

The IAD’s management interface usually consists of a PC equipped with software that allows the network manager to define and monitor traffic flow, bandwidth requirements, access line quality, and various configuration parameters.

Through this interface, administrative functions, such as the creation of call profiles, are also performed. A call profile is a file that contains the parameters of a particular data call so that a similar call can be quickly reestablished at another time simply by loading the call profile. The call profile function usually includes a factory-loaded profile that acts as the template for creating and storing user-defined call profiles. Because each data call may involve as many as 25 separately configurable parameters, the use of call profiles can save a lot of time. Users typically load or edit a call profile via the management system using keyboard commands.

Inverse multiplexer management interfaces often support remote devices. This capability allows a network administrator at a central location to configure, test, and otherwise manage other inverse multiplexers at remote locations in much the same way as is currently offered by the in-band management systems of some T1 multiplexers. This is accomplished by the management interface reserving a certain amount of the network bandwidth, usually not more than 2%, as a subchannel to implement remote management.

Most inverse multiplexers can be remotely monitored and controlled via SNMP. This is usually accomplished with SNMP agent software included with the product. The agent collects detailed error statistics, utilization ratios, and performance histories that can be retrieved for analysis.

A solid base of standards now exists to allow equipment vendors, service providers, and end users implement a wide range of applications via ATM. The standards will continue to evolve as new applications emerge. The rapid growth of the Internet is one area where ATM can have a significant impact. With the Internet forced to handle a growing number of multimedia applications—telephony, videoconferencing, faxes, and collaborative computing, to name a few—congestion and delays are becoming ever more frequent and prolonged. ATM backbones have played a key role in alleviating these conditions, but other technologies such as MPLS and packet over SONET (PoS) allow next generation networks to be used to their full potential.

As noted in Chapter 5, MPLS selects specific routing paths based on source identifiers and the requested destination. MPLS is becoming the preferred protocol for enabling QoS in IP networks, while PoS offers a backbone architecture that preserves existing investments in SONET infrastructure and supports the deployment of IP-based video and voice applications. With PoS, the IP layer is placed directly above the SONET layer, eliminating the need to run IP over ATM. This enables QoS guarantees but dispenses with the overhead required to run IP over ATM over SONET.

---