GPRS PERFORMANCE MANAGEMENT

AND OPTIMIZATION

Issue Date: August 2005

Abstract—This paper discusses performance management and optimization schemes for mature GPRS
networks. A selection of GPRS KPIs based on network performance counters, drive tests, and active measure-
ment systems, as well as possible threshold values and their analysis and interpretation, is presented. Ongoing
and problem-triggered optimization and fine-tuning, including considerations regarding network design and
configuration, are also presented.

INTRODUCTION data services triggered the implementation of a

oday, although wireless local area network packet-switched (PS) overlay network, opening
T (LAN), worldwide interoperability for
microwave access (WiMAX), and associated
the public land mobile network (PLMN)
operators to the market of data networks and
technologies can provide a radio access network Internet-based services. Voice services, however,
(RAN) for Internet Protocol (IP)-based services in remain the core business for many European
distinct areas, these hotspots are a long way from PLMN operators, who introduced the tendency
providing contiguous coverage. Therefore, many to optimize the existing infrastructure from a CS
wireless operators are focused on providing perspective. This tendency and other factors have
mobile Internet access. A large variety of contributed to the growing requirement to
technologies has contributed to the operators’ optimize GPRS.
vision of true mobility and seamless roaming. This paper focuses on the performance
The currently prevailing technology providing management and optimization of GPRS
true mobile Internet access on a European scale is networks. A performance management overview
general packet radio service (GPRS). New is presented, including proposed key
generations of mobile handsets, personal digital performance indicators (KPIs) for the RAN as
assistants (PDAs), and BlackBerry® devices, as well as the core network (CN). A combination of
well as GPRS-enabled PCMCIA1 product cards, network counter and protocol analysis, as well as
are enjoying increasing market penetration. The drive testing, is introduced as a reliable basis on
dominance of GPRS networks highlights the which to isolate problems. This is followed by a
importance of taking full advantage of discussion of optimization techniques based on
capabilities. It remains to be seen whether the the configuration of GPRS signaling procedures,
ongoing deployment of third-generation (3G) the usage of available network features, and the
access networks will at some point not only performance considerations for IP backbones.
supplement existing GPRS coverage in Europe,
but replace it.
GPRS PERFORMANCE MANAGEMENT
Cellular operator success was built on circuit-
erformance management involves capturing
switched (CS) services. The popularity of short
messaging and the growing demand for mobile
P and analyzing a set of KPIs that can be
quantified by using different sources of
____________________________ performance data. Network operators typically
Dirk Michel define a set of KPIs and stratified target
1 PCMCIA is the name of an international trade association
dmichel@[Link] and standards body (Personal Computer Memory Card thresholds in line with marketing and business
International Association) and is also used to describe the priorities. Performance management teams—
PC cards themselves (small form factor devices used in a
which are responsible, for example, for certain
variety of applications and devices).

© 2005 Bechtel Corporation. All rights reserved. 15

 ABBREVIATIONS, ACRONYMS, AND TERMS
3G third generation MNC mobile network code
APN access point name MRTG Multi Router Traffic Grapher
ARQ automatic repeat request MS mobile station
ATM asynchronous transfer mode MSC mobile switching center
BCCH broadcast control channel multi-RAT multi-radio access technology
BDP bandwidth delay product N-cell neighbor cell
BECN backward explicit congestion NMS network management system
notification NTE network termination
BER bit error rate equipment
BLER block error rate (P)AGCH (packet) access grant channel
BSC base station controller PBCCH packet broadcast control channel
BSS base station system PCU packet control unit
BSSGP BSS GPRS protocol PDA personal digital assistant
BTS base transceiver station PDCH packet data channel
CCCH common control channel PDP packet data protocol
C/I carrier-to-interference (ratio) PDU protocol data unit
CN core network PLMN public land mobile network
CPU central processing unit (P)RACH (packet) random access channel
CS circuit switch(ed) PS packet switch(ed)
CSD circuit switch domain PSD packet switch domain
DE discard eligible QoS quality of service
DHCP dynamic host configuration RADIUS remote authentication dial-in user
protocol service
DLCI data link connection identifier RAN radio access network
DNS domain name service RAND random number
FECN forward explicit congestion RAU routing area update
notification RLC radio link control
FPDCH fixed packet data channel RTT roundtrip time
FR frame relay Rx/TxQual receive/transmit quality
FRoATM FR over ATM RxQual receive quality
FTP file transfer protocol SGSN serving GPRS support node
GGSN gateway GPRS support node SI system information
GPRS general packet radio service SIM subscriber identity module
GSM global system for mobile SNMP simple network management protocol
communication
SQI speech quality index
GSN GPRS support node
SRES signed response
GTP GPRS tunneling protocol
TBF temporary block flow
HLR home location register
TCH traffic channel
IMEI international mobile
equipment identity TCP transmission control protocol
IMSI international mobile TMSI temporary mobile subscriber identity
subscriber identity UMTS universal mobile telecommunications
IP Internet Protocol system
IPoA IP over ATM UTRAN UMTS terrestrial radio access
network
ISP Internet service provider
VC virtual circuit
KPI key performance indicator
VLR visitor location register
LAN local area network
WAN wide area network
LAU location area update
WiMAX worldwide interoperability for
LLC logical link control microwave access
MCC mobile country code

16 Bechtel Telecommunications Technical Journal

 Um

BTS

Abis Network
Counters
Um

BSC/ MSC
RAN HLR
PCU (VLR)
A D The number of
Gc packet access
Gb Gs
BSS
Network requests and the
Gr Counters
Test Mobile transmitted data
Stations for IP Core volume during
SGSN GGSN Internet
Active Testing
Gn Gn Gi CS busy hour
Network Gp can be used
Counters
to determine
External
whether the
GGSN
GPRS CS resource
Network
requirements are
Signaling and Data Transfer Interface
“pushing” the
Signaling Interface
PS traffic into
congestion or
Figure 1. Measurement Points
whether the GPRS
districts of major cities, towns of a certain size, or GPRS KPIs traffic demand is
large rural areas, as well as routing areas or IP Selecting meaningful KPIs is a precursor to
backbones—periodically monitor various KPIs to continuous performance monitoring. GPRS KPIs
genuinely high.
isolate areas not achieving target performance. should encompass the base station system (BSS)
Optimization schemes are then initiated to ensure and the CN to account for end-to-end
that performance targets are met. performance, i.e., from the MS to the Gi interface
on the gateway GPRS support node (GGSN).
KPIs are commonly established using
Typically, KPIs for the BSS and the RAN reflect
performance counters provided by network
accessibility, retainability, and integrity. CN KPIs
elements. Statistics based on network element
focus on routing area update (RAU) behavior,
counters can be verified and supplemented by
congestion metrics on several interfaces, and
additional measurements obtained by active
system accessibility. Table 1 provides an
testing with test mobile stations (MSs), protocol
overview of GPRS KPIs.
analyzers located at different interfaces in the
network, and deployed active performance GPRS BSS KPIs
measurement systems. Such additional testing • BSS Accessibility
systems can highlight aspects of GPRS The RAN plays a prominent role in overall
performance not evident from network counters system performance, since it typically
alone, such as duration times for several signaling represents the bottleneck in terms of
procedures that can affect the user-perceived resources and available physical
system quality. Figure 1 presents the GPRS transmission rates on the air interface.
network architecture and possible measurement Accessibility from a RAN/BSS perspective is
points for network counters and protocol mainly a function of radio resources,
analyzers. Standard network surveillance or signaling capacity, and BSS parameter
alarm monitoring, as well as long-term trending configuration. Packet channel requests on the
and analysis of multisource KPIs, can assist in (packet) random access channel ([P]RACH),
identifying problems. which constitute the first contact of an MS

August 2005 • Volume 3, Number 1 17

 Table 1. KPIs and Thresholds for GPRS System Performance

NETWORK NETWORK
ATTRIBUTE KPI
ENTITY ELEMENT

BSS BSC/PCU Accessibility (P)RACH

BSS BSC/PCU Accessibility Immediate Assignment Rejection Rate ([P]AGCH)

BSS BSC/PCU Accessibility Congestion Time

BSS BSC/PCU Accessibility PDCH Allocation Failure Rate

BSS BSC/PCU Accessibility TBF Setup Failure Rate

BSS BSC/PCU Quality 4 out of 4 Success Rate (Multislot Class)

BSS BSC/PCU Quality TBFs per PDCH

BSS BSC/PCU Quality PDCHs per TBF

BSS BSC/PCU Integrity RLC BLER

BSS BSC/PCU Integrity RLC Retransmissions

BSS BSC/PCU Quality Cell Reselection Success Rate

BSS BSC/PCU Quality Cell Reselection Time

BSS BSC/PCU Retainability Percentage of TBF Preemption

CN SGSN Accessibility GPRS Attach Success Rate

CN SGSN Accessibility GPRS Attach Time

CN SGSN Accessibility PDP Context Activation Success Rate

CN SGSN Accessibility PDP Context Activation Time

CN SGSN/GGSN Quality GPRS RAU Success Rate

CN SGSN/GGSN Quality GPRS RAU Success Time

CN SGSN/GGSN Quality Congestion Levels on Gi, Gb, and Gn

System End-to-End Quality Delay

System End-to-End Quality Application Throughput

BLER block error rate GPRS general packet radio service (P)RACH (packet) random access channel
BSC base station controller (P)AGCH (packet) access grant channel RAU routing area update
BSS base station system PCU packet control unit RLC radio link control
CN core network PDCH packet data channel SGSN serving GPRS support node
GGSN gateway GPRS support node PDP packet data protocol TBF temporary block flow

with the network, are transmitted via slotted congestion; the immediate assignment
ALOHA2 [1, 2]; unsuccessful access requests rejection rate is based on the number of
due to collisions are therefore intrinsic to the immediate assignment rejections over the
access method. Increased levels of (P)RACH total number of (P)RACH requests. Such
request failures, however, can indicate poor rejections are due mainly to the lack of packet
radio conditions. The total number of data channel (PDCH) resources in the cell,
(P)RACH requests may include instances although excessive delays on the Um
where the MS sends multiple consecutive interface or lack of resources in the base
access bursts. Immediate assignment—the station controller (BSC), packet control unit
BSS’ response to the packet channel request (PCU), or other nodes may cause rejections [3].
from the MS—is pivotal for identifying
Cell congestion levels can be calculated as the
____________________________ time of zero PDCH availability within the
GPRS busy hour, which may be expressed as
2 Slotted ALOHA is a technique devised in 1972 that
a percentage of 60 minutes. GPRS congestion
doubled the capacity of ALOHA (a system developed a few
years earlier at the University of Hawaii to coordinate and levels should also be interpreted in the
arbitrate access to shared communication channels). context of traffic channel (TCH) demand for

18 Bechtel Telecommunications Technical Journal

 CS connections. On the one hand, CS busy the other hand, it should be noted that GPRS
hour traffic can affect available resources for busy hour patterns are not necessarily
PS connections and obviously depends on congruent, but are offset to those of the
the BSS configuration in terms of preemption circuit switch domain (CSD). It may,
and the allocation of fixed PDCHs. The therefore, be useful to consider CS traffic
number of packet access requests and the (expressed in Erlangs) and PS traffic
transmitted data volume during CS busy (expressed in bytes) busy hour statistics
hour can be used to determine whether the when assessing GPRS congestion. For
CS resource requirements are “pushing” the illustrative purposes, sample traffic profiles
PS traffic into congestion or whether the have been provided in Figure 2. Figure 2a
GPRS traffic demand is genuinely high. On presents CS and PS traffic distributions based

Routing Area (1 Month)

120 1600
PS (Data Volume) [MB]
CS (Erlang) [E] 1400
100

1200
Data Volume (MB)

80
1000

Erlang [E]
60 800

600
40
400

20
200

0 0
O:00 2:00 4:00 6:00 8:00 1O:00 12:00 14:00 16:00 18:00 20:00 22:00
Time of Day

Figure 2a. Sample Traffic Profile – Routing Area

Network-Wide (1 Month)
9000 160,000
PS (Data Volume) [MB]
8000
140,000
CS (Erlang) [E]
7000
120,000
Data Volume (MB)

6000
100,000
Erlang [E]

5000
80,000
4000
60,000
3000

40,000
2000

1000 20,000

0 0
O:00 2:00 4:00 6:00 8:00 1O:00 12:00 14:00 16:00 18:00 20:00 22:00
Time of Day

Figure 2b. Sample Traffic Profile – Network-Wide

August 2005 • Volume 3, Number 1 19

 on 4-week averages for a routing area are subscriber-specific, unidirectional, logical
containing 120 cells. Each point on the graph connections associated with a varying
represents the average value calculated number of PDCHs, established to facilitate
across 20 weekdays during a particular hour data transfer between the MS and the
of the day. The average total data volume BSC/PCU [3]. In other words, the data
from 24:00 to 01:00 is 39.61 MB, based on service may be accessed successfully, but the
20 values at the same hour. The figure shows connection quality can range from
that both CS and PS distributions are unacceptably bad to very good.
correlated and that their peaks fall into the
The PCU reserves the requested number of
same period (17:00 to 18:00). The voice curve
PDCHs for new connections during the
exhibits the expected pattern, with a
packet assignment procedure. PDCHs
secondary peak at lunchtime and the primary
available in the packet switch domain (PSD)
peak close to 18:00, followed by a rapid
at any given moment are used for the
decline in traffic. The PS curve shows a
assignment, although more than one TBF
similar trend up to the primary peak, from
may be assigned to a PDCH. The GPRS
which the traffic levels are sustained for
multiplexing capability can support multiple
approximately another 3 hours before
TBFs per PDCH and allocate new TBFs on
declining. This GPRS traffic curve resembles
already active PDCHs. The absolute or
those of typical Internet service providers
average number of TBFs on a single PDCH is
(ISPs). Figure 2b provides a network-wide
typically governed by system parameter
reference graph for the same period of time to
settings, which also regulate when new
verify the traffic patterns shown for a
requests for on-demand PDCHs can be
randomly chosen routing area, although the
issued. On-demand PDCHs are requested
primary peak for CS and PS falls between
when the limit of allowable TBFs per PDCH
16:00 and 17:00.
has been reached, which is captured by the
The PDCH allocation failure rate (sometimes TBF-per-PDCH KPI. Note that the number of
referred to as GPRS blocking) and the PDCHs requested by an MS is determined
temporary block flow (TBF) setup failure mainly by its multislot class. On-demand
rate are additional KPIs that can flag PDCHs make temporary use of idle CS TCHs,
congestion situations. A certain amount which are transferred back to the CSD when
of GPRS blocking may be acceptable they have not been used for a defined period
considering that CS-centric cell dimensioning of time.
strategies may allow for approximately
Additionally, fewer than the supported
2 percent blocking or quality of service (QoS).
number of PDCHs for a given multislot class
QoS levels may vary, however, depending on
may be assigned to the MS if the total number
the operator’s preference.
of requested PDCHs cannot be provided due
to failed on-demand and fixed packet data
Interfaces
channel (FPDCH) requests. The successful
Gb interface between the SGSN and the BSC/PCU
assignment of all requested PDCHs per
Gi interface between the GGSN and the multislot class can be monitored by the
public data network
4 (or 3) out of 4 (or 3) success rate (multislot
Gn interface between the GSNs class) KPI. When active resources are released
Gr interface between the SGSN and the HLR by other connections, TBF upgrades may be
Gs interface between the SGSN and the MSC initiated to allow the allocation of the PDCHs
Um air interface between the MS and the BSS to TBFs that have fewer resources than those
that provides circuit and packet data supported by the MS multislot class.
services over the radio interface to the MS
New GPRS users can be blocked if the PSD
cannot be expanded to accommodate new
TBFs when the TBF-per-PDCH limit has been
• BSS Retainability
reached, resulting in PDCH allocation
RAN CS congestion, genuinely high volumes failures (packet access reject) and TBF set-up
of data traffic, or a combination of both can failures. Note that PDCH allocation failures
give rise to a series of other symptoms that may occur after successful immediate
may affect not only the system accessibility assignment. These failure rates can be biased
but also the retainability and quality for by TBF upgrade failures. System vendors
ongoing TBFs, i.e., GPRS connections. TBFs may provide a feature that allows the TBF to

20 Bechtel Telecommunications Technical Journal

 request additional PDCHs if it has been set Desirable RLC retransmission rates are
up initially with fewer PDCHs than the generally below 1 percent.
multislot class of the MS allows.
Additionally, TCP window sizes, TCP
Although TBF multiplexing and the retransmission time-outs, typical IP packet
allocation of fewer resources than the MS can size distributions, and the application
support work together to provide flexible protocol affect the effective throughput on
system accessibility, both mechanisms can the application layer [5, 7, 8]. For a detailed
have an undesirable effect on throughput analysis of higher layer performance over
performance and bandwidth delay product GPRS, refer to the aforementioned references.
(BDP) per user. Additionally, the absolute or
• BSS Quality and Mobility
average number of allocated PDCHs per TBF
in a given cell provides another perspective Cell reselection, as well as possible location
and routing area updates (LAUs and RAUs), The PDCH
on GPRS QoS. Multiplexing is indicated
when this KPI increases beyond 1. Because has a deteriorating effect on throughput allocation failure
the logical link control (LLC) throughput performance. During cell reselection, the MS rate (sometimes
performance per TBF is affected by the releases all channels to read the system
referred to as
multislot class of an MS, the distribution of information of the target cell. The radio
outage time begins with the channel release GPRS blocking)
different multislot MSs in the network should
be identified to set achievable throughput and ends with the completion of the cell and the TBF
targets. Analyzing Gb interface log files can reselection process and the establishment of setup failure rate
provide the dominant multislot class MS. downlink and uplink TBFs on PDCHs of the
are additional
target cell. Cell reselection may also require
• BSS Integrity LAUs and RAUs if the target cell is part of KPIs that can
The varying nature of radio propagation and another location and routing area. This flag congestion
RAN interference conditions are the main introduces extra signaling sequences with the situations.
factors affecting data integrity in the BSS. The network as discussed below and increases the
carrier-to-interference (C/I) ratio has a time until new TBFs are established on the
critical impact on the maximum physical target cell. It should be noted that the radio
transmission rate [4, 5] of a PDCH. The GPRS outage also affects the higher layer
coding schemes, whose ranges of optimum throughput. The corresponding IP outage
performance are related to the C/I ratio, help time begins with the last received IP
to reduce the impact of radio interference on datagram on the source cell and ends with
the radio channel’s block error rate (BLER). the first received IP datagram on the target
Remaining uncorrectable bits and corrupted cell. During this outage time, the network
radio blocks trigger the radio link control buffers arriving datagrams, which introduces
(RLC) automatic repeat request (ARQ) increased delay times and decreased BDP.
mechanism if the TBF is operated in The number of cell reselections and
RLC-acknowledged mode [3]. The proba- associated procedures are directly correlated
bility of RLC retransmissions thus reduces to the mobility of the MS [9].
the physical transmission rate of the channel,
GPRS CN KPIs
adds additional delays to the air interface,
The GPRS CN plays a vital role in assessing
and affects the end-to-end roundtrip time
overall system performance. Among the key
(RTT) of higher layer protocols [4, 5].
elements of CN performance assessments are
Increased delays due to RLC retransmissions,
success rates for GPRS attach, packet data
however, appear to affect the quality less
protocol (PDP) context activation, RAU, and
than operating RLC in unacknowledged
paging. The average amount of time required to
mode and leaving possible retransmissions to
perform the above procedures may also be
the upper protocol layers, bearing in mind
relevant when determining system performance,
that ARQ on the RLC and the transmission
although it should be noted that this depends
control protocol (TCP) do not pose an
heavily on the feature set used by the operator.
increased risk of protocol inefficiencies [6].
This makes it very difficult to directly compare
The bit error rate (BER) of a radio channel
approximate GPRS attach times and other delays
(also categorized into receive/transmit
among operators. Variations among system
quality [Rx/TxQual] levels ranging from
vendors may also exist. Benchmark values
0 to 7) and the percentage of RLC retrans-
provided by system vendors (dependent on the
missions should be closely monitored.
software and hardware revision of the network

August 2005 • Volume 3, Number 1 21

 nodes) can be used. However, long-term trending Gb interface congestion and dimensioning
of such procedure completion times can help should be considered when assessing the
identify critical network variations. GPRS attach success rate. The Gb interface
interconnecting the BSS with the SGSN is a
• CN Accessibility
frame relay (FR) connection over an E1/T1
The GPRS attach request message is the first link [11, 12]. Depending on the CN design
contact of the MS with the serving GPRS strategy, this FR connection may be
support node (SGSN) after a TBF has been transported over intermediate networks,
established successfully. It contains the such as asynchronous transfer mode (ATM)
temporary mobile subscriber identity (TMSI), backbones supporting FRF5 (FR over ATM
mobile network code (MNC), and mobile [FRoATM]). FR congestion control and
country code (MCC), as well as location and discard mechanisms, i.e., forward and
Common causes routing area identity. Subsequent identity backward explicit congestion notifications
of PDP context checks of the international mobile equipment (FECNs and BECNs), as well as the discard
identity (IMEI) or the international mobile eligible (DE) flags, can be monitored to
activation failure
subscriber identity (IMSI), as well as identify congestion levels on the FR data link
include network authentication and ciphering request connection identifier (DLCI) carrying the
link failures, messages, may be issued by the SGSN. To BSS GPRS protocol (BSSGP) protocol data
incorrect DNS retrieve the IMSI if no entry is found for it, units (PDUs).
the SGSN uses the old location area
records, and Measuring the PDP context activation success
information to identify the old SGSN where
inappropriate this terminal was being served last. The rate is equally important. The MS may
SGSN may require home location register initiate a PDP context activation upon
configuration
(HLR) signaling over the Gr interface to successful completion of the attach
of MSs. procedure. The MS initiating the PDP context
identify profiles for unknown IMSIs.
Unsuccessful authentication and ciphering activation request message provides certain
may also contribute to the attach rejects if the information to the network for an activation
operator has chosen to activate those features attempt to be successful [13]; PDP type, PDP
for attach procedures. The SGSN address type, access point name (APN),
authenticates the GPRS mobile by sending a password, QoS profile, and various protocol
random number (RAND). The subscriber options can all be defined in the request
identity module (SIM) applies GSM message. The SGSN receiving the activation
algorithms and the authentication key (Ki) to request issues a domain name service (DNS)
this RAND to obtain the ciphering key (Kc) query to identify which GGSN IP address
and the signed response (SRES), which are holds the appropriate routing table entry to
then sent back to the SGSN in the the requested APN. The SGSN routes the
authentication response [10]. If successful, context activation request to the identified
the Kc is used to cipher the payload. GGSN, which authenticates the GPRS
subscription against a remote authentication
KEYS dial-in user service (RADIUS) server [13].
Similar to a regular dynamic host
Kc ciphering
configuration protocol (DHCP) client, the
Ki authentication
GGSN typically follows with a dynamic PDP
address allocation request to a dedicated
SGSN counter statistics for the Gb interface DHCP server. Static and dynamic allocation
can be used to establish the probability of of IP addresses can be supported, although
attach failures and the distribution of failure dynamic allocation is generally preferred by
causes. Some of the common failure causes many GPRS operators. The use of external
are attach requests from users who are IPv4 and IPv6 address space can also be
not subscribed to the GPRS network, supported. The GGSN provides the allocated
international roamers from networks without IP address to the SGSN via the create PDP
roaming agreements, software and hardware context response, which, in turn, sends the
levels on specific MSs, attach time-outs activate PDP context accept message to the
(excessive delays, failed responses from the MS. Thus, a successful PDP context activation
MS), and network failures. Typical network establishes IP connectivity between the MS
failures include lower layer link errors, and the GGSN, which then forwards or
as well as HLR signaling failures on the routes the traffic to external networks.
Gr interface.

22 Bechtel Telecommunications Technical Journal

 Common causes of PDP context activation updates. The nonperiodic update requests
failure include network link failures, can be triggered when the MS transcends a
incorrect DNS records, and inappropriate routing area boundary during cell
configuration of MSs. The latter can lead to reselection. The intra-SGSN RAU procedure
authentication failures on the RADIUS server is used when the new routing area of the
and failed access requests to unknown APNs. target cell is administered by the same SGSN;
Counters on the SGSN and GGSN provide the inter-SGSN RAU is used when it is not.
the basis for establishing the PDP context
During RAU, the SGSN can commence the
activation failure rates and the distribution of
LAU procedure with the mobile switching
the causes. Like the above Gb interface
center (MSC)/visitor location register (VLR)
performance assessment, congestion levels
(known as a combined RAU) if the optional
and dimensioning shortfalls can be identified
Gs interface is used by the operator. If the Gs
on the Gn interface. In contrast to the FR
interface is not deployed, the MS itself is
interfaces used on the Gb interface, the Gn
required to issue the LAU request message
interface typically uses IP interfaces over
on the new cell before requesting RAU.
100-Base-Tx or ATM/AAL5 STM-1/OC3 to
Similar to the attach procedure, the LAU may
connect to the GPRS backbone network. The
require the MS to respond to IMSI or IMEI
IP backbone can be based on different
checks as well as authentication and
network architectures, although meshed
ciphering requests. From the time the MS
ATM platforms with IP over ATM (IPoA)
receives the channel release message after
interworking are typically used for
LAU completion until the MS issues the
converged, QoS-aware, high-capacity
channel request for RAU, the MS remains in
backbones. When a PDP context is in active
idle mode to read the system information
state, GPRS tunneling protocol (GTP)
(SI) 13 on the broadcast channel of the target
encapsulation is used to create virtual circuit
cell. Consequently, the MS needs to reacquire
(VC) connections between GPRS support
a channel to initiate the RAU, which may, in
nodes (GSNs) supporting the data flow
turn, require authentication and ciphering.
across the backbone. The GTP VC connection
RAU rejects can result from authentication
is released as soon as the context becomes
and ciphering failures, RAU time-outs due to
inactive. Various ATM performance counters
excessive delays, lower layer link failures,
of the Gn interface are provided by the SGSN
and/or protocol failures.
and GGSN nodes, which typically include
ATM adaptation-layer-related counters as Figure 3 summarizes the GPRS KPIs and the
discarded AAL5 PDUs, and ATM layer- elements that affect them.
related counters as cell loss ratio, cell loss
priority cells, and others. ATM provides a KPIs for End-to-End System Performance
multitude of congestion control schemes that Aside from the ability to access the GPRS
should be part of any ATM network design to network, one of the outstanding KPIs is the
avoid congestion and accomplish recovery. achieved application throughput between the MS
Additionally, operators can introduce and the host. Application-specific throughput
commercial network management systems performance is also one of the hardest to quantify,
(NMSs) that provide real-time APN-specific bearing in mind that network counters do not
metrics for attach time, attach success rate, provide direct quantifications for this KPI. One
and context activation time. Such systems are possible way to monitor application throughput
often provided with Web interfaces and can is to deploy active performance measurement
help to quickly assess system performance. systems. Such systems often require that
stationary MSs be installed and client hardware
• CN Quality and Mobility be distributed among the routing areas as well as
The RAU success rate forms another key the server platforms at a central location. The
metric for GPRS system performance. client typically has protocol analyzer facilities
Intelligence about this KPI contributes to the and configurable functions that enable
understanding of end user throughput consecutive attach cycles, context activations,
performance. A GPRS-attached MS can issue different application protocols, and predefined
periodic, intra-SGSN and inter-SGSN RAUs, traffic profiles. Performance metrics can be
depending on its situation. Periodic RAUs used for GPRS attach, context activation,
are used to ensure that the MS is still authentication time, DNS response time, TCP
reachable and are thus treated as intra-SGSN connection establishment time, and additional

August 2005 • Volume 3, Number 1 23

 BSS

Using
GPRS-specific
broadcast
channels and
algorithms is
a fundamental CN
decision that
many European
operators have
not yet made.

Figure 3. KPI Impact Matrix

application-specific indicators. Figure 4 provides GPRS RAN/BSS

a possible installation scenario for an active The RAN plays a prominent role in GPRS
performance measurement system. optimization. Coverage planning, frequency
planning, and neighbor cell definitions have an
obvious impact on radio conditions and
GPRS NETWORK OPTIMIZATION interference levels, which should be carefully
PRS network optimization schemes can vary controlled. The RAN can aid in understanding CS
G from operator to operator and may be tied to
specific circumstances, although certain practices
retainability and handovers of a given cell when
assessing radio conditions. Interference levels can
and approaches have a more universal character. deteriorate receive quality (RxQual), speech quality
On the one hand, optimization should be index (SQI), and other CS-related KPIs. Frequency
understood as an ongoing activity in the context retunes and neighbor cell corrections often
of network rollout and expansion, and one that is improve high percentages of intra-cell and
concerned mainly with an optimal integration of imperative inter-cell handovers or TCH drops and
new base stations, BSCs, and CN elements. The handovers due to downlink quality. A “clean”
RAN, however, more than any other part of the radio environment is a prerequisite for GPRS
GPRS system, changes constantly as new cells performance due to its impact on possible
are introduced and existing ones are transmission rates of the air interface (Um)
decommissioned, relocated, or otherwise [4, 5, 6, 7]. Adequate channel dimensioning
modified. Optimization teams also support strategies of base transceiver stations (BTSs) for
network operations on an ongoing basis. On the CS and PS services should ensure minimum
other hand, optimization activities may be congestion levels.
triggered when certain areas of the network fail to Supported BSS features and parameter
meet performance targets. Performance reviews optimization can further improve system
and subsequent optimization can be used to performance. Using GPRS-specific broadcast
target specific shortfalls. On the whole, GPRS channels and algorithms is a fundamental
optimizers are generally grouped into RAN/BSS decision that many European operators have not
and CN teams.

24 Bechtel Telecommunications Technical Journal

yet made. A principal concern is the fact that not A general decision is also made in the context of
all MSs in the operator’s network support packet CS and PS prioritization or preemption.
broadcast control channels (PBCCHs). Roaming can Depending on GPRS priority and preemption
often pose a compatibility issue, and a dedicated parameters, on-demand PDCHs can be treated
Um time slot is required for GPRS signaling. as available or blocked for CS connections,
regardless of a PDCH’s current usage state.
An important feature that can be provided by PS
Generally, priority and preemption algorithms
broadcasts, however, is network-assisted cell
are affected by using dynamic allocation or
change, which allows a considerably reduced
adapting half-rate channels for CS connections,
radio outage time during cell reselection. When in
but the implementation can vary among
packet transfer mode, the MS sends packet cell
system vendors. Operators typically allow the
change notifications to the BSS before entering
preemption of on-demand PDCHs to allocate
the cell reselection procedure. The network’s
new CS connections when needed. This Cell reselection
response contains minimum required system
approach, however, begins to conflict with a success rates and
information about the target cell, which allows
baseline performance guarantee for GPRS users
the MS to initiate TBF connections in the target completion times,
in congested situations.
cell before cell reselection is actually completed. including their
Where TCH requirements are high and PDCH
The PBCCH also enables the operator to use impact in terms of
preemption causes GPRS blocking (packet
GPRS-specific cell plans. Typically, GPRS cell radio outage and
immediate assignment rejects) to be 2 percent or
reselection decisions are controlled by CS idle
higher during the CS busy hour, a strategy for IP outage, can be
mode parameters because the MS remains in idle
allocating FPDCHs can improve accessibility established by
mode during packet idle and packet transfer
performance. FPDCHs may not be preempted, so
modes. The MS behaves, therefore, as it does in active drive
the defined number of time slots is guaranteed for
CS idle mode, even though it is GPRS attached. testing in
GPRS users. Additionally, if GPRS blocking is
GPRS-specific cell plans allow the independent
2 percent or higher during the GPRS busy hour, the field.
optimization of GPRS cell reselection in packet
the allocation of FPDCHs can be considered.
transfer mode, i.e., reselection is not governed by
FPDCHs can also increase throughput
CS parameter settings. Prominent idle mode
performance during blocked and congested
parameters include cell reselection offset and
situations. Software-based simulations and test
hysteresis, penalty timers and penalty offsets, and
trials on the production network can be used to
minimum required receive levels for accessing
identify the optimal number of FPDCHs for a
the system.
target GPRS congestion time, target throughput,
The possibility of prioritizing cells for GPRS packet size, TCP window size, and MS multislot
traffic is another feature available with GPRS- class. Target values for immediate assignment
specific signaling. A behavior similar to CS reject rate and TBF setup failure rates are
hierarchical cell structure can then be used to typically between 1 and 2 percent. It should be
“move” traffic from certain cell types to others. It noted, however, that the allocation of FPDCHs
can be desirable, for example, to prioritize cells of reduces the number of Um time slots available to
indoor schemes, thus minimizing traffic caused the CSD, introducing the potential for degrading
by overlapping macro cells. the CS service. Ensuring an acceptable level of

AMS Fixed BTS BSS

Data Center Client Hardware,
Store Including MS
Server

NTE Um
Gb
Gn
WAN SGSN
IP
Backbone GGSN
SGSN Gn
Server Hardware NTE BTS Gn
Logically Connected
to Client Hardware Gb
AMS
Server Um
AMS Fixed
Client Hardware,
Including MS
BSS

Figure 4. Sample Active Performance Measurement System

August 2005 • Volume 3, Number 1 25

 GPRS performance without significantly affecting the number of allocated PDCHs that carry a
the CS service becomes part of the operator’s TBF provider should be analyzed. Excessively
FPDCH allocation methodology. long release times lead to reduced capacity
utilization rates.
Cell reselection success rates and completion
times, including their impact in terms of radio
outage and IP outage, can be established by active GPRS CN
drive testing in the field. Network counters do not GPRS CN design and supported features make a
necessarily provide statistics that can be used to critical contribution to overall system performance,
establish the duration of cell reselection and especially in the context of processes with an end-
outage when in GPRS packet transfer mode. to-end character. A clear prerequisite for CN
Typically, the number of PDCH requests, TBF performance is the analysis of link utilization and
requests, and other counters are provided on a central processing unit (CPU) load of the involved
The Gs interface network interfaces and nodes, including those of
per-cell basis. Identifying cell reselection
can be useful performance on a statistically representative basis the IP backbone. CN capacity should be monitored
to reduce is, therefore, challenging; carefully selecting a regularly to accommodate the growth of GPRS and
certain number of sample cells in adequate universal mobile telecommunications systems
the amount of
locations in different routing areas can help (UMTS) subscribers. In multi-radio access
signaling between technology (multi-RAT) networks, GPRS- and
to identify anomalies in the RAN. The
the network and measurements can be retrieved by drive trials UMTS terrestrial radio access network
the MS, because with test MSs performing bulk downloads using (UTRAN)-originated traffic can be supported by the
file transfer protocol (FTP). Tracing cell reselection same CN platform, in which case it is imperative
it allows location
signaling messages as well as data transfer that regular dimensioning audits be performed.
and paging Ideally, simple network management protocol
activities in the source cell and target cell provides
information to be the desired information. Because of the impact of (SNMP)-enabled devices should be used to monitor
passed directly reselection, usage of cells with small coverage the IP backbone for packet loss and delay, as well as
footprints should be minimized in areas where link utilization, which can be achieved with SNMP-
between the
high levels of mobility are expected, especially for enabled devices. Multi Router Traffic Grapher
SGSN and (MRTG) [14] and SmokePing [15] are two
motorway and railway coverage. It should be
the MSC. noted that MS type and software level affect cell commercial tools3 that can help to achieve
reselection times, since these autonomously real-time trending. Typical RTT values in GPRS
trigger cell reselection. Another element that networks vary between 700 and 1200 ms. Packet
affects cell reselection times is the availability of losses of 1 to 5 percent are common.
access grant blocks on the common control GPRS attach, as well as many other access and
channel (CCCH), which is used by the BSS to issue change request operations, can require the MS to
immediate assignment messages. SI13 broadcast perform various security checks. Attach
cycles can vary depending on the BCCH procedures can be configured with a varying
configuration. Shorter SI13 broadcast intervals number of security features, including identity
can, therefore, reduce the amount of idle time check, ciphering, and authentication. These
during cell reselection that the MS uses to read features can be implemented with different
the SI13 of the target cell when a PBCCH requirements (e.g., constant check, selective
is not present. check, systematic check, and random check) that
Delayed TBF release should also be considered. regulate when the network issues requests to the
When TBFs are operated in acknowledged mode, MS. Because it is generally desirable to reduce the
uplink and downlink TBFs are released when all number of signaling handshakes and thus the
RLC blocks have been received and acknowledged time needed to complete GPRS attach, operators
by the MS or the BSS. It may, however, be desirable must identify the level of security required.
to retain TBF connections during temporary PDP context activation after successful GPRS
interruptions of data transfer, thus avoiding attach can also introduce substantial delays.
ongoing TBF releases and establishments. Identifying context activation times with protocol
Increasing this delay, however, may lead to the testers on the Gn interface can reveal if a
inefficient use of radio resources, because the
probability of a TBF remaining established without
____________________________
traffic being sent also increases. Typical delay
values are between 2 and 3 seconds. PDCH 3 Multi Router Traffic Grapher and SmokePing are
utilization can be useful to determine the freeware, freely available under the terms of the GNU
General Public License.
effectiveness of the configured release time, i.e.,

26 Bechtel Telecommunications Technical Journal

particular message exchange is causing excessive
delays or failures. RADIUS and DHCP servers, MESSAGES
including the available IP address space per APN, ACK acknowledge
should be dimensioned appropriately. Typical SYN synchronize
RADIUS setup success rates should be greater SYSACK system acknowledge
than or equal to 95 percent, and setup times
should range between 20 and 150 ms. The CPU
load of the servers and the most common failure
causes should also be noted. Operators can CONCLUSIONS
choose to initiate GPRS attach on MS after
PRS system performance relies on
power-on, thus reducing the user’s perception of
the time required to establish an IP connection.
G performance management schemes, which
identify performance shortfalls, and on network
One of the basic CN design capabilities is the optimization, which is conducted continuously
implementation of the optional Gs interface. The Gs and also when triggered by problems. GPRS
interface can be useful to reduce the amount of performance from the terminal equipment to the
signaling between the network and the MS, host is a function of the BSS as well as of the CN;
because it allows location and paging information thus, optimization activities should focus on both
to be passed directly between the SGSN and the areas. C/I and BLER, radio capacity, PCU and
MSC. The LAU signaling procedure between GSN dimensioning, RADIUS, DNS and DHCP,
network and MS is not required, consequently mobility, usage of BSS and CN features, and
decreasing RAU times as well as the IP outage time. configuration all contribute to the end-to-end
performance experienced by the user. Active
The impact of the current SGSN configuration on measurement systems can be used to accurately
LAU and RAU completion times can be capture application throughput and RTTs.
measured by appropriate protocol analyzers on Network operators should ensure that global
the Gb interface, as indicated in Figure 1. The system for mobile communication (GSM) and
individual signaling messages can be traced to GPRS networks are balanced in terms of
derive statistically meaningful standard message resourcing performance management and
transfer times and anomalies or outliers, optimization, acknowledging the increasing
accordingly. Conducting such measurements on importance of GPRS in the context of
all available Gb interfaces in the network helps to supplementing IP services over UMTS.

establish target times for LAU and RAU
completion. Due to the different signaling
procedures involved in intra- and inter-RAU,
TRADEMARKS
differing RAU success rates are to be expected.
Differentiating between intra- and inter-RAU BlackBerry is a registered trademark of Research
success rates can help isolate possible In Motion Limited.
performance issues between neighboring routing
areas, without a possible bias due to intra-RAUs.
REFERENCES
To assess the IP outage time, active field tests are
[1] 3GPP, TS 03.64 V8.11.0 (2003–04), Technical
required, since network counters typically do not Specification Group GSM/EDGE Radio Access
provide session-specific performance data. Test Network; General Packet Radio Service (GPRS);
MSs engaged in bulk downloads from an FTP Overall Description of the GPRS Radio Interface;
Stage 2 (Release 1999).
server located on the intranet, including TCP/IP
[2] 3GPP, TS 44.060 V6.10.0 (2004–11), Technical
protocol analyzers and tracers installed on the Specification 3rd Generation Partnership Project;
terminal equipment, can increase understanding of Technical Specification Group GSM/EDGE
IP outage times. End-to-end packet loss, TCP Radio Access Network; General Packet Radio
Service (GPRS); Mobile Station (MS) – Base
retransmission ratios, and delay measurements Station System (BSS) Interface; Radio Link
provide additional metrics to characterize Control/Medium Access Control (RLC/MAC)
throughput performance. The time required to set Protocol (Release 6).
up TCP connections is another indicator; the [3] 3GPP, TS 04.60 V8.23.0 (2004–05), Technical
Specification Group GSM/EDGE Radio Access
number of issued synchronize (SYN), system Network; General Packet Radio Service (GPRS);
acknowledge (SYSACK), and acknowledge (ACK) Mobile Station (MS) – Base Station System (BSS)
messages per connection should be close to one. Interface; Radio Link Control/Medium Access
Control (RLC/MAC) Protocol (Release 1999).

August 2005 • Volume 3, Number 1 27

 [4] T. Halonen, J. Romero, and J. Melero, GSM, ADDITIONAL READING
GPRS and EDGE Performance – Evolution Towards
3G/UMTS, John Wiley & Sons, 2nd Edition, 2003. • 3GPP, TS 23.060 V6.3.0 (2003–12), Technical
Specification Group Services and System Aspects;
[5] R. Chakravorty and I. Pratt, “Performance Issues General Packet Radio Service (GPRS); Service
with GPRS,” Journal of Communications and Description; Stage 2 (Release 6).
Networks (JCN), Vol. 4, No. 2, December 2002,
pp. 226–281.
[6] S. Hoff, M. Meyer, and A. Schieder, “A BIOGRAPHY
Performance Evaluation of Internet Access via
the General Packet Radio Service of GSM,” Dirk Michel joined Bechtel
Vehicular Technology Conference (VTC) ’98, Telecommunications in 2003.
May 1998, Ottawa, ON, Canada. He is currently based in the UK
[7] X. Chen and D. Goodman, “Theoretical Analysis as a senior RF engineer,
of GPRS Throughput and Delay,” IEEE overseeing the configuration,
International Conference on Communications integration, and optimization
(ICC) 2004, June 2004, Paris, France. of GSM/GPRS cell sites. He is
also responsible for network
[8] D.E. Comer, “Internetworking with TCP/IP –
Principles, Protocols, and Architectures,” performance analysis and
4th Edition, Prentice Hall, 2000. characterization of GPRS.
Before joining Bechtel, he worked for AOL Europe
[9] D. Michel and N. Ramasarma, “GPRS (Luxembourg) and Star21 Networks (Germany). With
Measurement Methodologies and Performance
AOL, he worked for the European Operating Centre,
Characterization for the Railway Environment,”
focusing on dial-in and DSL networks. At Star21, he
Wireless Communications and Networking
Conference (WCNC) 2005, March 2005, New worked as a fault management team leader and
Orleans, LA. network planning and development engineer. His
responsibilities included ATM and IP core network
[10] G. Heine, “GSM Networks: Protocols,
design, traffic engineering, and capacity planning.
Terminology, and Implementation,” Artech
He was also part of the R&D team at Star21
House Publishers, 1999.
Laboratories, responsible for product acceptance
[11] 3GPP, TS 04.64 V8.7.0 (2001–12), Technical testing, hardware compatibility testing, and the staging
Specification Group Core Network; Digital area for network rollout.
Cellular Telecommunications System (Phase 2+);
General Packet Radio Service (GPRS); Mobile Dirk received a Diplom Geograph degree in
Station – Serving GPRS Support Node Geographical Information Science from the Justus
(MS-SGSN) Logical Link Control (LLC) Layer Liebig University, Giessen, Germany. He is a member of
Specification (Release 1999). the Institute of Electrical and Electronics Engineers
[12] 3GPP, TS 48.016 V6.1.0 (2004–11), Technical Communications Society and the Institution of
Specification Group GSM EDGE Radio Access Electrical Engineers.
Network; General Packet Radio Service (GPRS);
Base Station System (BSS) – Serving GPRS
Support Node (SGSN) Interface; Network
Service (Release 6).
[13] 3GPP, TS 07.60 V7.2.0 (2001–03), Technical
Specification Group Core Network; General
Packet Radio Service (GPRS); Mobile Station
(MS) Supporting GPRS (Release 1998).
[14] MRTG ([Link]
[15] SmokePing ([Link]
smokeping/[Link]).

28 Bechtel Telecommunications Technical Journal