Three Hidden Dangers of GPRS Deployment White Paper
P/N 340-1278-001 REV B
White Paper
Three Hidden Dangers of
GPRS Deployment and
How to Avoid Them
March 2004
Spirent Communications, Inc.
26750 Agoura Road
Calabasas, CA
91302 USA
Email: productinfo@spirentcom.com
Web: www.spirentcom.com
North America
+1-800-927-2660
Europe, Middle East, Africa
+33-1-6137-2250
Asia Pacific
+852-2166-8382
All Other Regions
+1-818-676-2683
Copyright
© 2004 Spirent Communications, Inc. All Rights Reserved.
All of the company names and/or brand names and/or product names referred to in this document, in particular, the
name “Spirent” and its logo device, are either registered trademarks or trademarks of Spirent plc and its subsidiaries,
pending registration in accordance with relevant national laws. All other registered trademarks or trademarks are the
property of their respective owners.
The information contained in this document is subject to change without notice and does not represent a commitment
on the part of Spirent Communications. The information in this document is believed to be accurate and reliable,
however, Spirent Communications assumes no responsibility or liability for any errors or inaccuracies that may appear
in the document.
Three Hidden Dangers in GPRS
Deployment and How to Avoid Them
Growing demand for wireless Internet data access is fostering double-digit growth in new
mobile data delivery technologies, especially GPRS. Indeed, GPRS subscriptions continue
to grow at a rate of over 30% quarterly.
1
This paper discusses specific benchmarking
recommendations for determining the performance of the core GPRS packet network.
2
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Hidden Dangers in GPRS Deployment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Preparation for Deployment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Looking Forward . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Spirent Communications White Paper | 1
1. Source: World Cellular Data Metrics – Quarterly Datasheet – December 2003, EMC, part of
Informa Telecoms Group.
2. Published in the 5th World Wireless Congress, San Francisco, May 25-28, 2004.
Three Hidden Dangers in GPRS Deployment and How to Avoid Them
Introduction
Introduction
GSM, the Global System for Mobile communications,
3
represents 70 percent of the
world’s wireless market, serving nearly 1 billion subscribers
4
in nearly 200 countries.
5
With such global coverage, GSM has become a leading focus for overlaying wireless data
services onto an existing service provider network. Today, that data service is
predominantly GPRS.
However, increasingly, operators are migrating to EDGE and even to UMTS while
maintaining or lowering prices. This presents an increasing strain on the core packet
networking as more users transfer more data more quickly and connect for longer periods
of time. This, in turn, demands carefully-planned benchmarks to ensure proper capacity
planning and robustness.
Hidden Dangers in GPRS Deployment
In spite of the attractions of GPRS and its successors, the sudden explosion in subscriber
populations coupled with new models for flat-rate v. content-based billing is threatening
both the robustness of the network and the operators’ power to charge for services that
they successfully deliver. In particular, there are three specific areas that demand special
attention and examination.
Danger #1: Unknown GGSN Performance Ceilings
While GPRS’s air interface throughput maximum of only ~150 kbps may seem
inconsequential, mobile subscriber populations could quickly reach tens to hundreds of
thousands. In such cases, even “low bandwidth” subscriber access links can quickly
overwhelm even a Gigabit Ethernet GGSN uplink.
Consider that most of the largest GPRS operators are reporting quarter-over-quarter
growth in double-digit percentages
6
– many exceeding 50%. Such explosive growth
challenges even the most diligent capacity planning efforts and thus precipitates an urgent
need for benchmarking for at least an order of magnitude greater subscriber populations.
Of particular concern is the concentration of subscribers on the GGSN. Figure 1 on page 3
depicts the many-to-one cascade of base stations (BS) to base station controllers (BSC) to
SGSNs to the GGSN.
3. Originally, GSM stood for “Groupe Spéciale Mobile,” the organization that spearheaded the
initiative. However, the group soon changed its name to “Global System for Mobile
communications” to give the new technology a more universal tone.
4. Source: GSM Association, January 2004 press release at http://www.gsmworld.com/news/
press_2004/press04_06.shtml.
5. Source: GSM Association, April 2003 statistics.
6. Source: World Cellular Data Metrics – Quarterly Datasheet – December 2003, EMC, part of
Informa Telecoms Group.
2 | Spirent Communications White Paper
Three Hidden Dangers in GPRS Deployment and How to Avoid Them
Hidden Dangers in GPRS Deployment
Figure 1. Many-to-One Cascade from Base Station to GGSN
Consider, too, that there is a cluster of mobile stations attached to each base station, and
that each mobile station will generate at least one GTP tunnel and may generate multiple
tunnels.
7
For this reason, the actual number of GTP tunnels to the GGSN may exceed the
number of active subscribers. Moreover, subscribers who are attached to the network but
inactive still consume resources on the GGSN, especially as those inactive subscribers
roam between SGSNs.
It is clear, then, that GGSN performance remains a critical factor in GPRS network
scalability, throughput, and availability
8
: The GGSN supports more GPRS connections
than any other device in the network and therefore occupies a central role in the
performance of the network. One could even argue that customer satisfaction – and
therefore customer churn – is increasingly dependent upon the GGSN.
7. Specifically, each mobile system will generate one or more primary tunnels and zero or
more secondary tunnels.
8. Eventually, UMTS will place even further demands on the SGSN. However, the current pre-
dominance of subscribers accessing the network via GPRS and EDGE places the burden on
the GGSN for the foreseeable future.
Spirent Communications White Paper | 3
Three Hidden Dangers in GPRS Deployment and How to Avoid Them
Hidden Dangers in GPRS Deployment
Benchmarking the GGSN’s performance to determine these ceilings requires that the test
environment:
Emulate thousands to millions of subscribers originating behind tens to hundreds
of SGSNs.
Exercise both steady-state and burst patterns for both bearer/data and control traffic.
Emulate both single and multiple primary tunnel configurations per mobile station,
and both zero and non-zero secondary tunnels per mobile station.
Include inter-SGSN roaming, validating that all primary and secondary tunnels
remain active during hand-off.
Figure 2 depicts a sample test of burst control and data plane activity. The filled triangles
indicate the number of active GTP tunnels (in hundreds) while unfilled circles indicate
instantaneous data throughput in kbps.
Figure 2. Sample Benchmark: Control and Data Plane Burst
1 The test begins with a few nodes attaching to the network and sending WAP data.
2 Once all GTP tunnels are active, data throughput reaches a steady-state plateau.
3 The test then initiates a burst of new GTP tunnels, simulating a wave of mobile users
attaching to the network.
4 Once all of subscribers attach, the test immediately initiates HTTP requests across all
of the active tunnels.
5 When all HTTP transactions are complete, the test begins to deactivate the more
recent group of GTP tunnels, simulating users detaching from the network.
6 Shortly thereafter, the test returns to the earlier steady-state (Step 2).
7 After a brief period, the test ceases the remaining data transfers and deactivates the
GTP tunnels.
4 | Spirent Communications White Paper
Three Hidden Dangers in GPRS Deployment and How to Avoid Them
Hidden Dangers in GPRS Deployment
Of course, this is merely one example of how control and data plane burst tests may
operate – in this case benchmarking the number of active calls and the aggregate data
quantity. Two other valuable performance metrics are GTP tunnel create (i.e., set-up) rate
and tunnel create time. That is, inasmuch as tunnel creation is not necessarily a serial
processes, the create time may not necessarily be the geometric inverse of create rate.
The graph in Figure 3 illustrates this point. GTP tunnels are created every 25 ms (i.e., at an
average creation rate of 40 GTP tunnels per second). However, each tunnel create process
takes 125 ms to complete. Thus, the time between successive tunnel creations is
independent of the time to complete a single tunnel creation.
Figure 3. Tunnel Creation Rate and Time are Unrelated
Of course, the same model may be applied to tunnel delete rate and time, as well.
Performance tests such as these enable accurate capacity planning through the knowledge
of the GGSN’s maximum performance and scalability.
However, beyond that maximum, the GGSN may exhibit different behavior in response to
periods of oversubscription. This brings us to the second danger.
Danger #2: Nonlinear Oversubscription Response
Considering the maturity of the installed GSM voice network, one might expect that (a)
GPRS data equipment is likewise capable of scaling to similar subscriber densities and (b)
that its responses to oversubscription are reasonably well understood. Unfortunately, such
response characteristics are seldom prescribed by operators. Indeed, some GGSNs may
crash and/or reboot in response to certain combinations of oversubscription conditions.
Spirent Communications White Paper | 5
Three Hidden Dangers in GPRS Deployment and How to Avoid Them
Hidden Dangers in GPRS Deployment
Determining how the GGSN responds to different types of oversubscription demands
benchmarking very specific oversubscription conditions, including:
Steady-state tunnel creation and/or deletion at rates that exceed the GGSN’s ceiling by
at least 20%.
Burst tunnel creation and/or deletion with a floor value near the expected high-water
mark combined with periodic bursts of 20% to 100% over-subscription.
Traffic bursts with a floor value near the expected high-water mark and periodic bursts
of 20% to 100% oversubscription. Such traffic is likely to be asymmetric.
Data/bearer traffic that includes both (a) a packet generation function to emulate
streaming data and to ensure continued oversubscription, and (b) real application
traffic to determine actual application response including possible session loss.
Such tests will prepare the operator for the network’s response during periods of
oversubscription. Furthermore, anticipating the likely severity and/or frequency of such
events is absolutely necessary for proper capacity planning.
This discussion thus far has focused exclusively on basic network performance and
response. This reflects the network pressures from emerging flat-rate mobile data
packages that are becoming particularly popular among the business sector.
However, operators are increasingly selling into the non-business “personal” mobile data
market with lower cost-of-entry packages coupled with “pay-as-you-go” options.
9
Such
endeavors have precipitated an emerging trend toward “content-based billing.” In this
model, mobile data pricing is based upon a laundry list of packages, applications, and
pricing strata. That brings us to the third danger.
Danger #3: Complex Application Tracking Requirements
The business case for per-application pricing models has already proven itself with today’s
plethora of charging models for messaging v. browsing v. traffic services v. news updates,
etc.
For instance, many operators assign per-transaction charges for MMS (e.g., picture-
messaging) based upon the expectation that an MMS message is likely to fall into one or
two generally predictable file sizes.
The following illustration depicts a network configuration in which (a) SMS (text
messaging) and MMS (picture messaging) are forwarded by the GGSN but not accounted
and (b) HTTP is accounted differently depending upon URL – either per transaction or per
kilobyte.
9. Pre-paid services fall under “pay-as-you-go” inasmuch as payment received prior to service
delivery is applied based on cumulative mobile data activity.
6 | Spirent Communications White Paper
Three Hidden Dangers in GPRS Deployment and How to Avoid Them
Hidden Dangers in GPRS Deployment
Figure 4. Sample Content-Based Billing Model
Similarly, many operators charge a per-message fee – albeit a lower one - for text
messaging
10
and instant messaging. By contrast, operators often charge per minute or per
byte for browser functions (e.g., WAP, HTTP).
If operators are to charge differently for different services, then they need a way to track
such activity accurately and a way to validate that accuracy. For instance, they may need a
method (a) to ascribe HTTP traffic per kilobyte to the subscriber’s account but (b) to
exclude SMS and/or MMS transactions since the latter will be charged separately and
tracked by the SMSC or MMSC.
Therefore, benchmarking the GPRS network means exercising the accuracy of accounting
and charging metrics. This means that the test application must:
Generate real application traffic (i.e., not merely protocol headers), including multiple
unique transactions per application (e.g., multiple URLs for WAP).
Measure number, duration, and payload per application and, where appropriate, per
content (e.g., per URL).
Exercise content-based billing features when the subscription density and/or the data
traffic load is near and above the capacity of the system under test.
Exclude network retransmissions (e.g., TCP window retransmissions) from ascription
to test results counters inasmuch as operators would usually not penalize subscribers
for the operator’s network anomalies.
Compare data throughput and subscription density per application when content-
based billing is enabled and disabled for that application.
11
10. Traditionally, text messaging relied upon Short Message Service operating on the SS7
signaling network. However, operators are increasingly migrating text messaging to data
services while maintaining the same user interface.
11. The operator must determine if accounting and charging for a transaction or application will
have a negative impact on the number of such transactions or applications that it can support.
Spirent Communications White Paper | 7
Three Hidden Dangers in GPRS Deployment and How to Avoid Them
Preparation for Deployment
Preparation for Deployment
The requirement for adequate performance, even during heavy network loads, creates
special challenges for the mobile data network: Service providers cannot risk offering
revenue-generating services unless they can first assess how effectively their systems will
deliver these services and how accurately they will track them during peak network loads.
Today, the mobile data industry lacks the necessary foundation of empirical data
describing the performance, scalability, and reliability of GGSNs in the face of exploding
traffic growth. Service providers are therefore taking significant – and unnecessary – risks
by deploying services in advance of such data.
The three dangers that we have identified here significantly increase the risks associated
with GPRS service delivery. Wireless data operators require accurate and scalable
benchmarking and validation prior to service provider deployment. Such tests must
include controlled and reproducible assessment of performance ceilings, network
availability during oversubscription and accounting accuracy.
Looking Forward
The benchmarks described in this paper presume that the operator has implemented
policies to ensure adequate performance of the data network. However, if the GPRS
network shares the same infrastructure as the voice network, then voice traffic may be
given precedence to transit lower-delay, lower-loss network routes and queues than data
services (such as GPRS). This puts GPRS at a disadvantage during periods of congestion.
Therefore, as push-to-talk and other packet-based voice services grow in popularity and
ubiquity, GPRS data services will warrant additional benchmarking in the presence of
competing voice-over-packet services.
Today, the recommendations described here will provide a solid foundation for
characterizing and designing reliable revenue-generating GPRS networks.
8 | Spirent Communications White Paper
