. 96
( 132 .)


rates, such as compressed video. Previously, video compression provided a constant
bit rate, sacri¬cing quality when compressing. Current compression techniques, such
as those offered by the MPEG schemes, maintain a ¬xed quality and instead provide
a variable bit rate. Thus the bandwidth requirements will increase when there is a
higher spatial and/or temporal resolution. VBR is further subdivided into two cate-
gories: real-time VBR (rt-VBR) and non-real-time (nrt-VBR). Non-real-time VBR is
traf¬c that is still classi¬ed as VBR, but may be buffered at the receiver, and thus
has much looser delay constraints. An example of this would be provision of a video
streaming service.
• Available bit rate (ABR): intended for traf¬c where the range of bandwidth requirements
is known. Applications which are bursty can be supported here. This guarantees a zero

loss as long as the application obeys the feedback from the network. In an ABR system,
the network will ask the application layer to slow down if there is congestion through a
feedback mechanism. Thus the ABR class is distinct from the other classes in that it is
inherently a closed-loop system. With such a system, it is possible to de¬ne a minimum
bandwidth that it will guarantee, and then a peak bandwidth that it will try to provide
if and when it is needed. However, it will make no promises with regard to this. For
example, the network may guarantee a minimum transfer rate of 2 Mbps, but with a
peak of 5 Mbps. A typical application using this service would be web browsing.
• Guaranteed frame rate (GFR): ABR traf¬c is inherently dif¬cult to provision for as
typically the peak data rate is not known or not relevant. Because of this, the ATM
Forum de¬ned a new type, GFR, to simplify the process. With GFR, a bandwidth level
is de¬ned such that traf¬c is guaranteed not to fall below this minimum bandwidth.
The traf¬c may receive a performance above this bandwidth, but at a best-effort per-
formance level. This service uses AAL5 and provides frame-level rather than cell-level
guarantees. This means that in a situation of congestion, a whole AAL5 frame, contain-
ing, for example, an IP packet, will be discarded. This provides a much more suitable
and ef¬cient service for transport of other protocols.
• Unspeci¬ed bit rate (UBR): provides no feedback or guarantees. The network will take
all cells travelling UBR, and transfer if there is any capacity. However, in the case of
congestion, they will be the ¬rst to be discarded and no information is sent back to
the sender. This is suitable for sending IP packets, as they do not promise delivery.
Typical applications would be email and ¬le transfer.

Figure 7.43 shows how the different traf¬c types are placed within the bandwidth of
the medium.
In UMTS, there is a de¬ned set of end-to-end QoS classes, which must be met by
any transport layer, including ATM. Table 7.17 shows the four service classes, their main
characteristics and how they map to ATM.





Figure 7.43 Traf¬c types

Table 7.17 UMTS service classes
Class Conversational Streaming Interactive Background
Delay Fixed, small Variable, small Variable Variable
Buffering No Yes Yes Yes
Symmetry Symmetric Asymmetric Asymmetric Asymmetric
Guaranteed Yes Yes No No
Adaptation layer AAL1/AAL2 AAL2 AAL5 AAL5
Example Phone call Video on demand Web browsing File transfer


Congestion occurs on networks for several reasons. One of the main ones is that data
traf¬c is often bursty by nature, i.e. ¬‚ow is not at a uniform rate. If traf¬c ¬‚ow was
smooth, then problems of congestion would be limited. To manage congestion, a traf¬c
shaping policy is needed to force the cells to be transferred at a more predictable rate.
When a user establishes a connection, it agrees to a traf¬c shaping for the transmission
with the network. Provided the user sends data in a way that conforms to the agreed
shaping, the network will play its part and promise to deliver all the cells. This reduces
congestion on the network.
ATM offers a number of key functions to support management of traf¬c on the net-
work, including:

• Connection admission control (CAC): the network policy and actions during a con-
nection setup which determine whether the connection should be admitted, rejected or
have its parameters renegotiated.
• Feedback control: ¬‚ow control mechanisms for the ABR class to maximize the band-
width usage and ef¬ciently share the available bandwidth among the users.
• Usage parameter control (UPC): network policing functions to ensure that negotiated
QoS parameters and constraints are adhered to.
• Traf¬c shaping: mechanisms implemented to ensure that QoS objectives are met by
the network and that traf¬c conforms to agreed parameters.

Unlike technologies such as Ethernet, which is a best-effort, non-deterministic scheme, in
an ATM network, the QoS mechanism allows a user to specify the type and level of service
required during the transmission. In essence, the user and the network enter into a contract
de¬ning the service. ATM allows that the terms of such a contract may, and probably
will, be de¬ned for an asymmetric link and therefore be different in each direction.
The provision of QoS is dealt with by the ATM layer, based on the QoS class that
has been de¬ned for the incoming virtual circuit or virtual path. The QoS is established
during the connection setup phase and relevant QoS parameters are passed through the
signalling messages. These parameters will provide a traf¬c descriptor and related traf¬c
parameters to characterize the QoS.

Before the connection can be admitted, each switch must not only ¬gure out which
output port can meet the service requirements, but also check if it can physically deliver
the resources by examination of what has already been allocated. This is the principle of
CAC, as illustrated in Figure 7.44.
To provide QoS across the network, each switch must also play a role to offer cells the
service class de¬ned for their connection. The key components of any QoS scheme are
mechanisms to ensure that data passes through with a certain delay, delay variation (jitter)
and loss characteristics. The method of implementation is through a classify, queue and
schedule (CQS) architecture. Essentially, a number of queues corresponding to different
service classes are established in the switch, and incoming traf¬c is classi¬ed and placed
in the appropriate queue. A scheduler then takes data out of the queues, processing the
queue with the highest priority, or level of service class, ¬rst, e.g. a voice call queue.
Ideally, to guarantee QoS, a switch would maintain a separate queue for each virtual
circuit that has been established. This mechanism is illustrated in Figure 7.45.
In a packet switched network environment, QoS offers only an approximation of the
performance, based on the information available at the time of connection establishment.
For some traf¬c types, such as voice, this will remain quite consistent for the duration of
the connection, but for others, may vary from the initial situation and may be adversely
affected by transient network events.

Can you
support this
ATM Switch



Figure 7.44 Call admission control

Classify Queue Schedule

port 1
port 2 port m
port n

Figure 7.45 CQS architecture

Policing Queue Schedule Switch Queue

port 1 port 1



port n port m

Figure 7.46 ATM switch CQS mechanism

CQS architecture is common in, and central to, ATM switches, but has not made a
signi¬cant impact yet to the IP protocol due to its best-effort nature. An IP router generally
has one queue and processes packets in a ¬rst in/¬rst out (FIFO) order. However, CQS
concepts will become increasingly important as IP introduces protocols to offer QoS, and
a major driver is the emergence of IP switching technologies such as multi-protocol label
switching (MPLS).
Enhancing Figure 7.45 to include more of the functionality of the system, Figure 7.46
shows the process cells are taken through as they pass through the switch. Each of these
steps is discussed in more detail in the following sections.

7.9.1 Traf¬c descriptor
The traf¬c descriptor speci¬es a set of traf¬c parameters that classify the ATM service
categories that were de¬ned earlier. To specify the QoS, these parameters are negotiated
between user and network, at connection establishment. The contract speci¬es the worst
possible value of the parameter and the network then guarantees to at least meet this.
The traf¬c descriptor parameters split into two general categories. Traf¬c parameters
pertain essentially to the speed and delay of transfer of traf¬c, and are what the user
de¬nes for the connection. The QoS parameters are tolerance levels that the user needs
for the transfer. There are also some non-negotiable network characteristics which are
not part of the traf¬c descriptor, but are rather measurements of the error performance of
the ATM network. Traf¬c parameters
These parameters govern the nature of the transmission in terms of speed and delay

• Peak cell rate (PCR): the maximum instantaneous rate at which the user will transmit.
The inverse of this is the time interval between cells. For example, if the user de¬ned
the interval between cells as 10 µs, then the PCR would be 100 000 cells per second,
i.e. 1/T. Figure 7.47 shows the cell interval.


1 2 3


Figure 7.47 Interval between cells

• Sustained cell rate (SCR): the average rate of cells measured over a long period. If, for
example, real-time uncompressed video is considered (CBR) then the SCR and PCR
will be the same, since there is no variation in rate of transmission. However, for bursty
traf¬c, the SCR may be low but the PCR high.
• Minimum cell rate (MCR): the minimum rate of cell transmission required by the user.
For an ABR service, the bandwidth will be between MCR and PCR and will most
likely vary a considerable amount between the two. However, it must never fall below
the MCR. A value of MCR = 0 would be a UBR service.
• Cell delay variation tolerance (CDVT): the level of variation in cell transmission times
that the application can tolerate. It can be considered as an error margin, and de¬nes an
acceptable level of deviation in cell transmission times. For example, a user generating
traf¬c at the PCR will need some margin of error as it is unlikely that they will be able to
guarantee that each cell will be sent with exactly the same time interval. Conformance
of this is measured in terms of the peak-to-peak cell delay variation (CDV).
• Maximum burst size (MBS): it is expected that for variable rate applications, they will
average at the SCR value, periodically bursting up to the MCR. The MBS de¬nes the
maximum number of cells that may be sent at the PCR. QoS parameters

These characteristics are as measured at the destination.

• Cell loss ratio (CLR): the percentage of cells that are not delivered to the destination.
This can be due to lost cells resulting from congestion or errors, or cells that arrive


. 96
( 132 .)