Asynchronous transfer mode
Asynchronous transfer mode (ATM) derives its name from the fact that the transmission system can accommodate traffic arriving at fixed time intervals, which is demanded by voice and video applications, as well as bursty traffic that arrives at no preset rate.
Multiplexers that maintain a predetermined relationship between the arriving traffic and its assignment to time slots on a shared transmission channel, called synchronous time division multiplexers, employ STM. In STM, the transfer is synchronized with the arrival of the traffic. The T-1 and E-1 carrier systems are examples of STM services.
Packet switching networks employ an entirely different technique—statistical multiplexing. In this situation, the occupancy assignment on the transmission channel is according to demand. The traffic is handled on a first-come, first-served basis. If traffic arrives and the channel is occupied at that moment, then the traffic is buffered and delayed until it can be transmitted. This works especially well in an environment where the information transfer is not delay-sensitive and it is bursty. Performance relies on the statistical arrival of the traffic.
ATM is a blend of these two systems. It has variable assignment, based on the arrival rate and delay sensitivity of the traffic. However, when the assignment occurs, it is to fixed-length time slots called cells. If the incoming traffic is delay-sensitive voice or video, then the assignment is immediate; however, if the arriving traffic is data, then it can be temporarily buffered before being transmitted.
ATM is akin to an escalator serving two types of traffic: people with infants who we assume should have priority access to the moving stairs and people who are able to wait. Steps of the escalators move at a constant rate, are a fixed size, and can accommodate only one person at a time. When the high-priority people arrive, they immediately walk onto the moving stairs, perhaps causing the low-priority people to momentarily stand aside. Once the priority traffic has been serviced, the other traffic can move onto the escalator. If there is too much high-priority traffic, then the escalator speed may have to be increased to keep the wait time within prescribed limits.
So What Is ATM?
ATM is a transmission scheme that combines aspects of statistical multiplexing and synchronous transfer mode. ATM provides the underlying technology for B-ISDN services. In ATM, data is organized into fixed-length entities called cells. Cells are delivered at a constant rate across the network. Whereas time slots have a fixed owner in a time division multiplexing (TDM) environment, cell ownership is not fixed in time. This is the asynchronous nature of ATM—cells have a constant bit rate but variable ownership.
Another important aspect of cell-based technology is that the small, fixed cell size results in a relatively low, controllable end-to-end delay for most applications. On the whole, cell relay is a form of packet switching; in fact, David Clark from MIT is credited with stating, “Cells ain’t nothin’ but small packets.” But the queuing delay suffered by a single cell is quite low compared to the propagation time across the network.
Consider the following example. The end-to-end propagation delay of a network might be about 10 ms. A packet or frame going through a typical switch will incur a delay of about 1–2 ms, or about 10–20 percent of the propagation delay. A single cell, however, will only experience a 10–20 µs delay, or a value that is 0.1–0.2 percent of the propagation delay.
Although the example above is a bit simplistic, it is meant only to make the point that queuing delays in a cell-based network are hardly noticeable compared with packet networks. It does, however, ignore the fact that multiple cells are necessary to create the same payload as a single packet/frame. Nevertheless, as long as everything is working well, cell delays are significantly lower than packet/frame delays.
ATM cells are 53 octets long. Although originally designed for high speeds such as 155 and 622 Mbps and beyond, early implementations have operated in the 1.544 to 100 Mbps range. Long-term plans from the ITU-T and ATM Forum already include rates well into the gigabit range.
ATM Service Characteristics
ATM provides an application-transparent service, which means the network does not know or care about the user information carried in the cells. Although ATM provides a connection-oriented and virtual circuit service, it can handle any type of user application. ATM can carry a variety of traffic types including voice, video, data, and images. Also it can provide the network infrastructure for (either public or private) local, metropolitan, and wide-area networks.
The ultimate goal of ATM networks is to provide support for a variety of services. ATM’s underlying switching infrastructure, however, is connection oriented (i.e., based on virtual circuits). Every cell header contains routing address information that references the virtual circuit. The term connection identifier is probably more descriptive of this virtual circuit address information. Whatever it is called, the cell header never contains an explicit endpoint address, as IP packets do.
Most connection-oriented networks that employ connection identifiers as header address information use them as a kind of “flat” space. That is, the identifiers are drawn from a series of numbers between some minimum and maximum, with a few always reserved or not allowed. For example, frame relay has a space that allows values from 0 to 1023; values less than 1000 can be used to identify user connections.
ATM connections are quite different. There are two levels of ATM connections: the virtual path and the virtual channel. They form a hierarchical relationship similar to the relationship between the IP network address and host address portions of an IP address.
Why are ATM connections different? The answer is that ATM is for everything. Telephones, PCs, and video cameras can all link over a public ATM network. In fact, the ATM network supports many different circuit types including ABR, UBR, VBR, CBR, and GFR. The number of connections required for end-user devices in a large office building with thousands of employees is staggering. The philosophy of ATM (and B-ISDN) is that a local, private switch could handle intrasite traffic switching and routing, while the public ATM network is concerned only with intersite switching and routing.
On the ATM network, the virtual path is for site-to-site connectivity, and the virtual channel is for device-to-device connectivity. Public ATM switches only need to look at the virtual path connection identifier to deliver the cell to the proper site. The local, private switch does the rest.
ATM Is Dead. Long Live ATM.
Studies suggest that telecommunications companies can significantly cut operating costs by moving away from traditional services and more to Ethernet-based offerings in the metro area. That, combined with a growing customer interest in Ethernet MAN implementations mean that carriers are beginning to shift focus from ATM, at least as a standalone service. However, surveys of ATM users also indicate that most of these customers have increasing budgets, so the potential revenue associated with such “mature” services cannot be ignored.
One gauge of ATM’s future is the switch market. IDC found that the market declined 12 percent in 2004 due to service providers moving to IP-based networks. And while they found there was sizeable demand because of DSL services, DSLAMs are also starting to be migrated to Ethernet platforms.
Initially, it might sound grim for ATM, but do old technologies really ever die? It appears that technology rarely is supplanted; rather, it is augmented and ultimately finds a new niche. Consider the fate of fire, the horse and buggy, or even the much-maligned Betamax video format! We still use fire for heat, cooking, and ambience even with the advent of central heating systems. Go to Central Park any day of the week and you will still find plenty of horse-drawn carriages. And while VHS won the consumer battle, Beta just might have won the war: contrary to popular opinion, Beta’s demise was greatly exaggerated as it found life in the TV broadcast industry and might just be around longer than VHS, which finds its market eroding because of things like DVRs, DVDs, and NetFlix.
Also consider that IBM has not gone away, despite the company's historical ties to the mainframe. Thomas J. Watson, Sr, wisely corrected his dubious assertion that the world would only need a handful of computers and ended up leveraging the existing card reader business by making it a front end to his company’s computer products. Now, “Big Iron” might not be selling much in new markets, but nobody’s throwing away their expensive old mainframes-they are instead using other technologies as the front end.
And to use a telecommunications example, what of private lines? People still buy T-carrier today (or at least DS-1s provisioned over HDSL and the like) despite the rise in frame relay's popularity. In fact, some companies are moving back to private lines because the prices have dropped dramatically and security/privacy concerns have grown in recent years.
It will likely be this way for ATM as well. Carriers have made significant investments in ATM over the years, and they are not just going to throw it all away. Much as they do with their POTS today, telecommunications companies will continue to make money from the infrastructure as long as they can.
ATM as Access to the Cloud
There was a time when the telecommunication industry anxiously anticipated the arrival of ATM. Once it arrived we would finally enjoy fully converged services, the sun would always shine, and world peace would spontaneously break out. There was a great deal of hype and hyperbole flying around about what ATM was going to do for carriers and customers, some of which was indeed fulfilled.
Yet a new technology has come to challenge ATM, and we are hearing many of the same stories breathlessly told again. This time, however, it might in fact be different. MPLS is being aggressively deployed and its beauty is that it does not require any particular Network Interface Layer (in TCP/IP language, the layer equivalent to OSI’s Physical Layer and Data Link Layer). So long as a customer supports IP, which pretty much everybody does, it can connect to the MPLS-enabled cloud.
However, the arrival of MPLS does not have to mean the death of ATM. Current ATM customers will be able to use the technology to access “the cloud” and more; the very nature of MPLS-enabled VPNs (based on RFC 2547bis) means ATM locations will be able to send IP data to frame relay locations, DSL locations, Metro Ethernet locations…truly communicate with any site no matter what network interface is available. Much like the Internet allows DSL customers to send email to cable modem users, provider IP networks will be the “glue” that ties many access methods together, including that dead old technology known as ATM.
|<mp3>http://podcast.hill-vt.com/podsnacks/2007q1/atm.mp3%7Cdownload</mp3> | Asynchronous Transfer Mode (ATM)|
|<mp3>http://podcast.hill-vt.com/podsnacks/2007q4/atm_vc.mp3%7Cdownload</mp3> | ATM VC Types|
|<mp3>http://podcast.hill-vt.com/podsnacks/2007q3/atm-cell.mp3%7Cdownload</mp3> | ATM cell size|