. 33
( 132 .)


illi s




Figure 4.49 Charging for roaming subscribers
4.9 CONNECTION MANAGEMENT 147 QoS mechanisms
Each ¬‚ow of data (PDP context) has a QoS associated with it, which de¬nes the follow-
ing classes:

• precedence
• delay
• reliability
• peak throughput
• mean throughput.

The QoS is requested during the activation but this may be modi¬ed by various elements
within the network in accordance with the resources available. The RLC/MAC layer sup-
ports four radio priority levels for signalling messages. The mobile device can indicate one
of these levels and whether the access request is for data or signalling. This information
is used by the BSS to determine the priority (precedence) of the information. The radio
priority level for user data is determined by the SGSN during the PDP context activa-
tion/modi¬cation procedures. The precedence class prioritizes signalling and data ¬‚ows.
Under normal operating conditions the network will attempt to meet the requirements
of all service commitments. However, under abnormal conditions it may be necessary
to discard packets etc. The precedence class priority level determines which ¬‚ows will
be maintained and which ¬‚ows will be affected ¬rst. There are three precedence classes,
with class 1 offering the highest priority and class 3 the lowest.

There are four delay classes de¬ned and these are outlined in Table 4.11. A delay class
of best-effort makes no guarantees about the delay experienced.

This is de¬ned in terms of residual error rates, which include probability of loss of data,
probability of data delivery out of sequence, probability of duplicate data and probability

Table 4.11 GPRS delay classes
Delay class Delay maximum value (s)
SDU size 128 bytes SDU size 1024 bytes
Average 95th Average 95th
transfer percentile transfer percentile
<0.5 <1.5 <2 <7
<5 <25 <15 <75
<50 <250 <75 <375
4 Best effort (unspeci¬ed)

Table 4.12 GPRS reliability classes
Reliability Modes of operation Traf¬c type
frame protection block
1 ACK ACK Protected ACK Non-real-time. Application
cannot deal with errors
2 UNACK ACK Protected ACK Non-real-time. Application can
deal with minimum errors
3 UNACK UNACK Protected ACK Non-real-time. Application can
deal with errors. GMM/SM
and SMS
4 UNACK UNACK Protected UNACK Real-time. Application can deal
with minimum errors
5 UNACK UNACK Unprotected UNACK Non-real-time. Application can
deal with errors

of corrupted data. Table 4.12, indicates the different classes and the appropriate modes of
operation to transport different types of traf¬c through the GPRS network.
Each different ¬‚ow of data will require speci¬c reliability classi¬cation according to
its particular requirements. The reliability classes state whether data is transferred in
acknowledged or unacknowledged mode across various lower-level protocols throughout
the network.

Throughput of used data is characterized in terms of bandwidth, split into a peak and mean
throughput class. As the name suggests, the peak throughput class de¬nes the maximum
rate at which the data is expected to be transferred. The class provides no guarantee that
the de¬ned rate will be reached, as this depends on both the resources available in the
network, and the capabilities of the mobile device. Table 4.13 shows the peak throughput
classes and the related data rate for each.
The mean throughput class de¬nes the average data rate in bits per second. The actual
de¬nition is in terms of bytes per hour, as the bursty nature of most data applications

Table 4.13 GPRS peak through-
put classes
Class Peak throughput (kbps)
1 8
2 16
3 32
4 64
5 128
6 256
7 514
8 1024
9 2048

Table 4.14 GPRS mean throughput classes
Class Mean throughput (bps) Class Mean throughput (bps)
1 Best effort 11 220
2 0.22 12 440
3 0.44 13 1110
4 1.11 14 2200
5 2.2 15 4400
6 4.4 16 11100
7 11.1 17 22000
8 22 18 44000
9 44 19 11100
10 111

means that although there may be high instantaneous demand for resources, on average
the throughput can be quite low. Table 4.14, shows the de¬ned classes. As with delay, a
class of best effort makes no guarantees. Traf¬c ¬‚ow templates
It is possible within GPRS to have a number of PDP contexts on a single device and
these can be connected via different GGSNs to completely independent networks. One
of these could be the Internet and another may be the subscriber™s home network. Thus
the mobile device will have two IP addresses, one for each context, and these are both
known as primary PDP contexts. Within R99 it is also possible to have two PDP contexts
that connect to the same network and share a single IP address. As an example, consider
that a subscriber may be connected to the Internet and simultaneously browsing the web
and downloading a large ¬le. When used in this method by sharing an IP address, the
second PDP context established is referred to as a secondary PDP context.
The traf¬c ¬‚ow template (TFT) is used to distinguish and separate packets arriving at
the GGSN from the same external network under more than one PDP context, but sharing
an IP address. The TFT will ensure that these are routed to the mobile device over the
correct tunnel for the PDP context. This enables the network to allocate different pro¬les
for each tunnel, to provide, for example, different levels of QoS or security.
Relating back to the example, by using separate contexts, the packets of data that are
associated with the interactive browsing session can be allocated a higher priority when
compared to the download session. Since the packets for both the browsing session and
the download session share the same IP address, the TFT is used to differentiate them
at the GGSN. The TFT is de¬ned using between one and eight packet ¬lters, which are
comprised of different sets of ¬elds within the IP header. The header attributes that may
be used are as follows:

• source address and subnet mask (since this is incoming from the external network,
the source is not the mobile device IP address, but rather the source address of the
incoming external IP packet)
• protocol number for IPv4 or next header for IPv6

• destination port range
• source port range
• IPSec security parameter index (SPI)
• type of service for IPv4 and traf¬c class and mask for IPv6
• ¬‚ow label for IPv6.

The above attributes are described in the IP description in Chapter 5. An example ¬lter
could be:
Packet Filter Identi¬er = 1
IPv4 Source Address =
Destination Port = 4000
Packets arriving at the GGSN from the external network with the above values can be
¬ltered by the GGSN and passed to the correct GTP tunnel associated with the speci¬c
PDP context

There are many factors governing how the core network is actually implemented. Each
has an impact on the QoS that the subscriber receives. The simplest scenario, illustrated
in Figure 4.50, shows a user wishing to connect to a home network which is in close
proximity to the mobile operator™s network. In this situation, the SGSN and GGSN may
be located in the same physical site, probably the same building. An Ethernet network
may be employed between the two units. However, if there is a large amount of data or
if a high level of QoS is required, then this could be some other high-speed technology
such as ATM or multi-protocol label switching (MPLS).
If there is a great deal of traf¬c between a mobile user and the home network, then a
leased line may be used as a connection between the GGSN and home router. This will
give maximum reliability and higher data rates but the cost will also be high. Another
option may be the use of a microwave link between the two sites. The initial expense

leased line Internet
BSS lea
IP over GGSN line
Ethernet or
geographically close

Figure 4.50 Simple connection scenario

may be high but the running costs of such a link can be very low, especially if it is using
the free-band on a point-to-point link and the power is kept below 1 W. This type of link
can give very high data rates at a fraction of the cost of a wired leased line. In many
cases, however, a great deal of error checking and error correction is required since these
radio links may not be so reliable.
If the user is in a different region to the home network, the connection implementation
has even more factors in the equation. It is also possible for lower traf¬c rates to use a
public network such as the Internet, or use a frame relay or ATM network. To introduce
security over the public Internet, it is likely that a virtual private network (VPN) will be
constructed between the mobile operator and the home Intranet (Figure 4.51). This will
probably be the most cost-effective solution and may be suitable for low volumes of data
transfers, which are not real time and can absorb delays.
An alternative would be to let the mobile network operator deal with the problem of
the subscriber being geographically remote. In this case, the SGSN (where the subscriber
is located) and the GGSN (where the leased line to the home network is located) may be
many hundreds of kilometres apart (Figure 4.52). In this situation, the mobile operator
has an estimate of how many roaming subscribers there are, and what sort of data rates
they expect. Since the SGSN and GGSN are geographically remote, Ethernet cannot be
used for this connection. The operator may use a public network such as the Internet to

leased line Internet
IP over

Ethernet or


. 33
( 132 .)