Unraveling ATM, part two
In our second look at asynchronous transfer mode, we investigate ATM service classes and reveal how they affect your network's quality of service and data traffic behavior
In this second part, we introduce the popular service classes constant bit rate, variable bit rate, unspecified bit rate, and available bit rate, and describe how the traffic parameters work. We will also take a look at how quality of service parameters affect ATM service. (2,500 words)
Last month, we began our definition of asynchronous transfer mode (ATM). Quickly summarized, ATM is a circuit-based, high-speed LAN and WAN technology based on synchronous optical network (SONET) fiber technology. It is circuit-based meaning that a clear circuit is defined between the source and destination computers which is actively monitored and maintained by all the intervening participant communications equipment. It uses high-speed SONET fiber technology to provide bandwidth rates in increments of approximately 52 megabits per second (OC-1), and all the way up to nine gigabits per second (OC-192). This technology works well for the LAN environment using copper wiring or multimode fiber cabling and for the WAN using single-mode fiber.
We discussed the difference between the circuit-based technology of ATM and the more traditional packet-based technology of today's IP routers. With circuit-switch technology it is possible to absolutely guarantee services for a connection without any aggregation or averaging of data delivery circumstances. Furthermore it is easier to deliver data through a pre-ordained circuit rather than having to examine each packet's header information and then pass it according to a look-up table. With this in mind, it is possible to deliver time-sensitive data such as streamed networked digital video on schedule as well as meter exactly for the amount of traffic delivered. On the other hand, circuit-switching technology can also be expensive since you have to incorporate much more intelligent systems into the equipment to keep the circuits performing as they should.
We last left off discussing the typical traffic parameters of an ATM circuit and the types of services that are affected by these parameters. The traffic parameters that we indicated in last month's column will come in handy here. These are the peak cell rate (PCR), sustained cell rate (SCR), and the minimum cell rate (MCR); I apologize for previously using minimum bit rate (MBR) as opposed to the more correct minimum cell rate (MCR). They contribute to how traffic flow is maintained in the various service classes we are about to describe.
ATM provides several classes of services with distinguishing behavioral differences depending on the type of information that is to be transmitted. The ATM specifications are defined by two major groups, the ATM Forum and the International Telecommunications Union (ITU). The ITU uses an alphabetic labeling of the classes. The ATM Forum has accepted the more popular, descriptive name-based labels. Keep in mind that the alphabetic class labeling of the ITU does not exactly match that of the ATM Forum for all of these service classes although there is close approximation. To avoid confusion, we adopt the more popular ATM Forum designations which is how most vendors describe their product capabilities. The classes listed are as follows:
Constant bit rate (CBR): CBR service provides a continuous time-synchronized stream of data traffic. This is most appropriate for time and loss sensitive applications such high-quality digital video transmission and private branch exchange (PABX) interconnects. This is an expensive option since the bandwidth is guaranteed between two points even if the line is idle; if no real data is being sent, the link will continue to send idle or blank cells at the same rate, effectively still using traffic when it really isn't necessary.
Real-time and non-real-time variable bit rate (rt-VBR and nrt-VBR): This service allows data to be delivered in bursts. The delivery of data may be synchronized, but it isn't necessarily guaranteed at a certain level. Using VBR, you can share the same link between more applications at the same time; this works best when applications have idle frames which could be used by other applications instead. VBR can be used for video conferencing where the loss of one frame or two won't really affect the result of the output significantly, especially when you consider that the typical TV signal runs between 30 to 60 frames per second. Real-time VBR provides for precise delay control between the two end points. Delay issues are not as critical for non-real-time VBR. Essentially one provides better quality levels of service than the other.
Unspecified bit rate (UBR): This class of traffic is what most of us experience on the Internet. The service makes no guarantees on delays or whether cells are delivered in sequence or at all. It leaves all these features up to higher-level protocols, such as the Internet Protocol (IP), to handle. When you loose cells, it is quite possible that an entire sequence of cells which constitute an IP packet (this can be a maximum of 64 kilobytes and consequently has to be chopped up to fit the many small 48-byte cell payloads), will be useless. The entire sequence of cells is tossed, and the higher-level IP protocol needs to signal the source that the whole packet must be resent. This procedure is known as early packet discard or tail packet discard and was added to UBR after its definition, leading some vendors to designate it as UBR+.
Available bit rate (ABR): This is similar to UBR in that it can vary the transmission rate of cells at any time. It differs in that better quality of service guarantees are available. ABR is what most hope will replace the UBR-like behavior of IP.
In addition to the traffic parameters, there is another group of parameters which describe the level of quality of service needed for a connection of each service class. Even among service classes, you can indicate what the maximum cell transfer delay (CTD), cell delay variation (CDV), cell loss ratio (CLR), and cell error ratio (CER) are allowed to be before any connection is terminated for unavailability at the rates and parameters indicated. Grouped together they are known as the quality of service (QoS) parameters.
The cell transfer delay is the amount of time that it takes for a cell to be taken from one end point to the far end point (with however many ATM switches in between). The cell delay variation is the maximum change of individual CTD amounts that are allowed. The cell loss ratio describes how many cells out of the total can be dropped in case of congestion. And finally, the cell error ratio describes how many cells out of the total can be dropped in case of transmission errors.
How these parameters affect service classes
The traffic parameters define the acceptable bandwidth rate and service class of a connection. The QoS parameters define how well the connection should perform. Not all parameters are used by all service classes; on the contrary -- there are other parameters which also affect these classes which are beyond the scope of our current discussion.
There are two ways to build a circuit when it comes to ATM. If you predefine a set of parameters and class and administratively create a set of circuits across your ATM network, you essentially build what is known as a permanent virtual circuit (PVC). This term may be familiar to those who have T-1 or fractional T-1 frame relay connections. A PVC is usually defined just once, at the time of installation of the circuit, and retains this type of circuit until you cancel your WAN service from your provider. On the other hand, a switched virtual circuit (SVC) is established when you run a specific network application and exists only for the duration that your application uses the services network. The network equipment (the ATM switch) builds a dynamic connection upon request with the given requirements.
PVC hardware is much easier and cheaper to build than that needed to support SVCs, since less intelligence is required on the switch to maintain connections. Although SVCs are a required part of the ATM specification, not all equipment can provide SVC connections; however, this is becoming less and less the case. You can have circuits of any of the service classes as PVCs or SVCs depending on how you wish to deploy a network. If, for example, you do not have any computers which have direct ATM connectivity, all you may need is a PVC. One of the problems with PVCs is logistical or administrative. Since most PVCs are configured manually, you have to either know ahead of time what the values for your traffic and QoS parameters should be or make educated guesses at them. Naturally more customers are looking forward to SVCs where the network performance of circuits are automatically determined by the switch, possibly saving significant costs in monthly traffic transmission, as well as providing more balanced and optimized services to network users.
CBR service typically has the same values for the PCR and SCR. The CLR, CTD, and CDV need to be defined per the requirements of the equipment that you will interconnect. High-performance digital audio or video transmission has much stricter and often higher demands than, for example, ordinary telephone calls between two sites.
Rt-VBR uses the PCR, the SCR, and another parameter called the maximum burst size, which is the total number of cells that are transmitted at the peak cell rate. This class usually results in traffic which averages around the SCR with small bursts at non-periodic intervals. Applications which are somewhat time-sensitive but can afford some degree of freedom or loss (such as video conferencing and even some audio applications) are best suited for this. The audio would be the equivalent of what it is like for good FM radio transmissions, where loss of one or two seconds occurs although rarely.
Nrt-VBR is similar to rt-VBR but is lower quality and therefore less suited for time-synchronous applications like video or audio. This would be better suited for pure connection-oriented data services. Frame relay over ATM is one good example of its use. It uses the same parameters, but only an average CTD and CDV are specified, rather than a maximum CTD.
UBR as we mentioned is similar in effect to the connectionless non-guaranteed data transfer mechanisms of the current Internet. With UBR the cell rate QoS parameters are not typically implemented; only the basic traffic parameters are used. To get any form of levels of quality in your network service, you have to properly architect your network. The SCR is specifically set to 0 since there are no minimum guarantees on bandwidth. The PCR is set to the maximum amount of bandwidth allowed on the connection; this upper limit is absolute according to the standard, and even if the physical link can afford larger bandwidth, the circuit will not go beyond the PCR.
And finally we get to ABR. This is right now the Holy Grail of ATM services when it comes to data networking. Many vendors have already implemented ABR in their equipment even though ABR is still being discussed in the ATM Forum and the ITU. ABR is another connectionless data transfer system except with provisions for QoS. Most of the future development is on Internet performance and on the next-generation Internet Protocol, IPv6 (Internet Protocol version 6, as opposed to the current IP version 4). Specifically there is concentration on how higher-layer protocols like IP can establish the QoS parameters of the lower-level ATM network.
ABR uses the minimum cell rate we indicated earlier. Effectively, the PCR, SCR, and MCR, identify a high, average, and low level that the bandwidth can be adjusted to on a dynamic basis. The ABR class looks quite a bit like the VBR service in concept; the difference lies in the fact that the extra unused bandwidth on the link is redistributed in a way that is more fair than that for VBR.
Go with the flow
Over the years, many have gotten used to the semi-unpredictable behavior of TCP/IP networks and the Internet that we have almost forgotten about acceptable service performance practices. As more and more companies join the Internet and the general congestion problem worsens, we can no longer stand to ignore these basic requirements. A new focus on providing quality services and guaranteeing uptime of network connections is becoming prevalent among service providers. The development of ATM technology, bringing not only connectivity services improved by magnitudes but also guaranteed services between points on the network, makes it a crux in the evolution of the global communications system.
We will continue our discussion on ATM technology and take a look at how vendors and their products are involved in the market. While focusing on IP networks, we will also take a look at how this ubiquitous protocol will behave in the new ATM order of networks.
About the author
Rawn Shah is vice president of RTD Systems & Networking Inc., a Tucson, AZ-based network consultancy and integrator. Reach Rawn at email@example.com.
If you have technical problems with this magazine, contact firstname.lastname@example.org