Networks today cope with multimedia traffic and can run out of bandwidth quickly. ATM networks increase data flow by intelligently controlling traffic and allocating different-size channels to different connections.
Mark Juliano
Networks grow in two ways: The number of attached devices increases, and the applications it supports become more diverse. The network administrator must maintain performance as traffic builds.
A shared-media LAN does not perform well under heavy traffic conditions or with a wide variety of traffic types. But ATM (Asynchronous Transfer Mode) overcomes these limitations by providing a connection-oriented approach to linking devices and by supporting a variety of traffic types, including audio and video. (See Peter Wayner's article ``On the Road to ATM,'' September BY
TE.)
ATM divides all traffic into fixed-length cells, each containing 48 bytes of user data and 5 bytes of overhead. Monitoring and organizing the flow of these cells through the network is called traffic management. How well it is done can be crucial, especially for time-sensitive data such as video.
My Dinner with ATM
To understand how to improve traffic flow in an ATM network, I'll compare the interaction of attached devices on a shared-medium LAN with the behavior of guests at a dinner party.
At the party, several people want to talk with one another; on a LAN, users want to send each other files and E-mail. For a dinner party to run smoothly, guests must behave according to a social protocol. Devices on a LAN must conform to a protocol so that they can communicate with each other. As the number of guests at a party increases, it becomes more difficult for an individual to speak with any other individual. So it is with LANs; as traffic increases, contention for access to the shared
transmission medium also increases.
Contention is the delay encountered while waiting for access to a shared resource, such as the Ethernet cable. Congestion occurs when network devices buffer their data until they can transmit it over a communications link.
Most LAN technologies share the physical transmission medium (e.g., copper wire and optical-fiber cable) among all the devices attached to the LAN. A protocol, or access method, organizes the sharing process that lets them use the medium. On Ethernet LANs, the access method is called CSMA/CD. The English translation is, ``Listen first, and if no one else is talking, one person may talk.'' Each network device first makes sure that no other device is transmitting and then sends its message. This type of LAN protocol is analogous to a small dinner party where four people are sitting and talking together. Only one person speaks at a time. As long as the number of people is small and no one talks all the time, everyone can take part.
Thing
s Can Change
As a business grows, its LANs expand. The more devices that are attached, however, the greater the transmission delays. Imagine a crowded restaurant trying to operate using the same rules as the small dinner party. Any person who wished to talk, whether it was with someone at their table or at the other end of the room, would have to wait until every other person in the restaurant was quiet before speaking.
This is why LAN performance often degrades as the number of attached devices increases; too many devices try to access the medium at the same time. Shared-media access methods are probabilistic; access is not guaranteed. There's no problem when the amount of LAN traffic is light (imagine a restaurant with customers at only one table), but a serious problem arises when traffic is heavy.
An Ethernet LAN addresses this problem by dividing the LAN into smaller, interconnected segments. (By analogy, the restaurant owner could build a separate room for each table.) While this reduc
es contention, it makes the LAN much more difficult to manage. (Imagine the manager of such a restaurant having to maneuver around--not to mention pay for--all those walls.)
ATM: An Alternative
ATM networks are connection-oriented and use a deterministic method to allocate network resources. Deterministic is the opposite of probabilistic; it guarantees access regardless of the number of devices attached to the network. The network is determined to get you access.
As each connection is established, the network allocates capacity to ensure that sufficient resources are available. ATM networks package traffic into cells of uniform length and manage the flow of these cells through the network.
Different types of traffic require different levels of service--referred to as QOS (quality of service). An ATM network provides different QOS levels to different traffic types. Indeed, different traffic types may call for different traffic management techniques.
Types of ATM Traffic
We can
classify the types of traffic supported by an ATM network according to three characteristics: bandwidth, latency, and cell-delay variation. Bandwidth is the amount of network capacity required to support a connection. Latency is the amount of delay associated with a connection. Requesting low latency in the QOS profile means that the cells need to travel quickly from one point in the network to another. Cell-delay variation is the range of the delays experienced by each group of associated cells. Low cell-delay variation means a group of cells must travel through the network without getting too far apart from one another.
ATM networks carry three types of traffic: CBR (constant bit rate), VBR (variable bit rate), and ABR (available bit rate). CBR traffic includes voice and video. To handle this traffic, the ATM network can act as a dedicated circuit. It provides a sustained amount of bandwidth, low latency, and low cell-delay variation.
VBR traffic is handled similarly to CBR except that the ba
ndwidth requirement varies. An ATM network supporting a videoconferencing application guarantees that a certain amount of bandwidth will always be available during a conference, but the actual bandwidth used can vary.
ABR traffic requires no specific bandwidth or delay parameters and is acceptable for many data applications. ABR connections support LAN traffic such as E-mail and file transfers. TCP/IP and NetWare will also use ABR connections.
The table ``Network Traffic Types and Their Requirements'' shows how link capacity is allocated to each traffic type. While CBR reserves a constant amount of the total available bandwidth, VBR requires that a large amount of spare capacity be available. ABR defines a way to use this valuable spare capacity. It can provide service that is no worse, and is in many cases better, than most of today's networks, but it doesn't require a constant bandwidth.
Making the ATM Connection
The ATM end station (the calling party) asks the ATM network for a conn
ection to another ATM end station (the destination) by initiating a connection request that leads to a negotiation with the ATM network. This process is called the connection establishment procedure. The parameters that must be negotiated are specified by the ATM Forum UNI (User-to-Network Interface) 3.0 standard and include traffic type, sustained and peak bandwidth, burst length, and QOS class.
This process secures a ``contract'' between the ATM network and the end station. The network promises to deliver a QOS, and the ATM end station promises not to send more traffic than it requested during the connection procedure. Such contracts must be enforced, and it is the job of traffic management functions to ensure that users receive the QOS they were guaranteed.
When congestion occurs, traffic management provides the mechanisms that allow the network to recover. ATM networks use three techniques: traffic policing, traffic shaping, and congestion control.
Traffic Policing
ATM networks ens
ure that traffic on each connection remains within the negotiated parameters. ATM switches use a ``leaky-bucket'' algorithm to police traffic. Imagine a wooden bucket with a hole in the bottom. Water leaks (traffic flows) out of a bucket (buffer) at a constant rate (the negotiated rate), regardless of how fast it comes in.
The need for police action occurs when traffic flow exceeds the negotiated rate and the buffer overflows. Each ATM cell header has a CLP (Cell Loss Priority) bit used to identify cells as either conforming (to the contract) or nonconforming. If cells are nonconforming--for example, more cells than the contract allows--the ATM switch sets the CLP bit to one. This cell can now be transferred through the network only if there is sufficient network capacity. If there is not enough bandwidth available, the nonconforming cell is discarded and may need to be retransmitted.
CBR traffic needs a single leaky bucket because it uses a sustained rate parameter in its network contract. VBR
traffic uses dual leaky buckets to monitor both the sustained rate over a discrete time period and the maximum (peak) bandwidth used during the connection. If either value exceeds the contract parameters, the ATM switch polices the VBR traffic by manipulating the CLP bit.
Traffic Shaping
Similar to traffic policing, traffic shaping is performed at the user-network interface. Traffic is manipulated so that flow rates obey the contract using a mechanism such as the dual-leaky-buckets algorithm. Devices that implement traffic shaping are typically ATM network adapters in PCs or workstations, hubs, bridges, routers, and DSUs (digital service units).
Congestion Control
Congestion can occur in any network, whether it uses shared media or ATM, whenever an application sends more data than the network can transmit with the available bandwidth. Also, as more applications send data over the same network, the bandwidth available to any one application changes over time.
Most networks fail to te
ll applications how much bandwidth is available at any given instant. As a result, applications have no basis on which to control the amount of data they send. When applications send more data than the network can handle, the network buffers fill up and can overflow. The application must then retransmit the data, which adds more traffic and further congests the network.
An ATM network performs congestion control so that ABR traffic can efficiently use the bandwidth that has not been guaranteed to CBR and VBR traffic. Effective congestion control reduces the need to retransmit data due to congestion.
While the problem of congestion control is still under discussion in the ATM Forum, it is expected that the final solution will use a variety of techniques, including end-to-end, link-by-link, rate-based, and credit-based traffic-flow control.
End-to-End vs. Link-by-Link. A network can control congestion over an entire connection path or by sublinks. With end-to-end control, the network measur
es the minimum available bandwidth along the connection and communicates the amount of bandwidth to the application, which then transmits at the appropriate rate. Each link in the network simply forwards the data as fast as it receives it.
Under the link-to-link technique, each link along the network connection controls its traffic flow independently. Each link buffers data as needed to adjust the incoming speed to the outgoing speed.
Rate-Based vs. Credit-Based. The rate-based traffic-flow technique constantly tells the application what transmission rate (the currently allowed rate) the sending device should use. For example, the network can tell an application that it can send 1000 cells per second. If the network becomes congested, it reduces this rate and notifies the application; when the network is no longer congested, the rate is increased.
The credit-based technique is slightly different. Here, the network indicates to the sending device the amount of buffer space (credits) availa
ble in the network. For example, the application may be allowed to transmit 100 cells, after which it must wait. The network periodically replenishes the application's credits. If the network becomes congested, the application gets fewer credits and they're replenished less often. This forces the application to slow the sending of data. When the congestion clears, the number of credits is increased and replenished fast enough that the application can transmit at full speed.
Integrated Congestion Control. The integrated proposal currently under consideration by the ATM Forum provides for an end-to-end, rate-based scheme as the default method, with the link-by-link scheme as an option where more precise control is needed. Because most existing ATM equipment already offers the default method, users should soon have a standards-based ABR congestion-control scheme.
If the network requires a more precise congestion management scheme, the link-by-link option can be used to control ABR traffic. When a c
onnection is made from an end-to-end device, the link-by-link device would simply perform the end-to-end flow control when talking to that device. This would preserve existing equipment, while providing for future growth.
Time to Switch?
No matter how carefully you plan a shared-medium LAN--whether Token Ring, FDDI (Fiber Distributed Data Interface), or Ethernet--the day will come when your network outgrows the capabilities of the technology on which it is based. ATM expands the network's ability to support a variety of applications that will help network administrators. Using ATM technology as a foundation ensures a widely accepted interface to the WAN and high-performance LAN, with traffic management designed to ensure that users get the quality of service they require.
Network Traffic Types And Their Requirements
TRAFFIC BANDWIDTH CELL-DELAY
TYPE EXAMPLE REQUIRED VARIATION LATENCY
Constant Voice,
Guaranteed Minimal Low
circuit emulation
Variable Compressed video Guaranteed Variable Low
Available Data Not guaranteed Widely variable Variable
Mark Juliano is vice president of marketing at Fore Systems, Inc., a supplier of ATM switches located in Warrendale, Pennsylvania. He can be reached on the Internet or BIX at
editors@bix.com
.
ATM Traffic Control
