Capacity planning in mobile data networks experiencing exponential growth in demand

Capacity planning in mobile data networksexperiencing exponential growth in demand.Informa’s 3G, HSPA & LTE Optimization Conference,17th April 2012, Prague, Czech Republic..

Dr. Kim Kyllesbech Larsen,

Technology, Deutsche Telekom AG.

The mega disruptive challenges …

Mega bits

Mega bucks

Mega Hz

2Dr. Kim Kyllesbech Larsen, 3G, HSPA+ & LTE Optimization, April 17th 2012, Prague, Czech Republic.

A typical data traffic day in Europe.

00:00 10:00 12:00 22:0017:00

Illustration

6:00 8:00

voicedata

@Work(2 – 4 Cells)

@Home(2 – 3 Cells)

On theGo

@Home(1 – 2 Cells)

On theGo

Small Cells


14:00

Today’s bit pipe and the bottlenecks.Network expansion as traffic “management” remedy.

Air/TRX/SiteSpectrum &Floor space

NodeProcessing

capacity

Backhaulbandwidth

Backbonebandwidth

CoreSwitching

WebApps servers

Bandwidth, CPU & Storage.

Off Loading(AP, Femto, …)

PacketCore

Web 2.0

+ Sectorization+ Small cells+ Additional spectral capacity (if available)+ Introduce more efficient technology

LL → MW → Fiber → + Colors

+ Colors+ switching

capacity

+ CPU+ switching

capacity

+CPU (i.e., CE, etc.)(up-to system limit)

RNC GGSNSGSN

traffic pressure pointsdue to aggregation

RNC GGSNSGSN

Node

+CPU

Optimized radio resource management (control plane)

PDP contextRABRRC


Data traffic trend to be considered.Most mobile data traffic is fixed-like in its usage.

50% of all traffic generated in 1 cell1.

80% data traffic carried by 3 cells1.

Remaining 20% carried over 28 cells.

50% of all traffic generated in 1 cell1.

80% data traffic carried by 3 cells1.

Remaining 20% carried over 28 cells.

Traffic off-load via WiFi & small-cell should be pursued more aggressively.

Traffic off-load via WiFi & small-cell should be pursued more aggressively.

31

2 3 4 3 3 2 2 2 1 1 1 1

0

5

10

15

20

25

30

35100% traffic80%+ traffic

Number of sites utilized perusage category.

1 on a per user basis. Note: This empirical law applies to volume as well as packet switched signaling.

“20% mobility”

Illustration


Law of small numbers of large consumption.Usage trend very much Pareto like.

12 month ago

80% subs took 20% of data traffic.

Today

A bit more than 30% of data traffic1

12 month ago

80% subs took 20% of data traffic.

Today

A bit more than 30% of data traffic1

Ca. 5% of active data users consume more than 1GB per month, more than 3 × the average monthly usage.

Ca. 5% of active data users consume more than 1GB per month, more than 3 × the average monthly usage.

1 Some of the diffusion over the 12 month might also be impacted by FUP cutting off the extreme usage.

Customers versusData Volumetric Consumption

Data Volumetric Consumption

Illustration


3G traffic distribution50% of sites carries 80% of 3G devices and 95% of 3G traffic.

Relative few network resources serves most of the demand.Relative few network resources serves most of the demand.

Illustration

0%

20%

40%

60%

80%

100%

0% 20% 40% 60% 80% 100%

3G Devices

3G-Traffic Volume

20% of 3G-cells carries 50% of 3G devices.




20% of 3G-cells carries 60+% of 3G traffic.

50% of 3G-cells carries 95% of 3G traffic.

20% of 3G-cells carries 60+% of 3G traffic.

50% of 3G-cells carries 95% of 3G traffic.

3G-Cells

@ Busy Hour 3G-Devices, 3G-Traffic


Postpaid trends – growth slowing down?

100%

Active PostpaidData users

“Basic phone”

SmartphoneiPhone

180%160%

80%

30%

50%65%

2009 2010 2011

Data growth

Smartphonepenetration

Volume growth

95%

35% 40%65%

120%

275%

2009 2010 2011

Data customer growth

Total

AndroidAndroid

Note: >90% of all smartphones are active data users. 65% of all postpaid have a smartphone, iPhone has a 40% share of all postpaid smartphones.

Android27%

Apple44%

Blackberry14%

Other15%

Illustration


Prepaid trends – the next growth wave!

100%

Active PrepaidData users

“Basic phone”

SmartphoneiPhone.

Note: 61% of all prepaid smartphones are active data users. Ca. 20% of all prepaid have a smartphone, iPhone share is 10% of all prepaid smartphones.

250%

500%

3% 7% 22%

2009 2010 2011

Prepay data growth

Smartphonepenetration

Volume growth

60%

170%

400%

550%

2009 2010 2011

Prepay data customer growth

Total

Android

Android19%

Apple10%

Blackberry55%

Other16%

Illustration


The difference between post- & pre-paid?

1 ….

Postpaid

Prepaid

00 02 04 06 08 10 12 14 16 18 20 22

Postpaid

Prepaid

Daily volumetric profile Busy Hour usage patterns

15 : 1

4 distinct postpay usage segments with 3 similar for prepay 4 distinct postpay usage segments with 3 similar for prepay

Illustration


OS …Last 12 month in smartphone heavy MNO.

1 Size of bubbles = share of active devices.

Illustration

Jan-11

+12 Month

“Basic phone”

Symbian

Windows

Volume development per device

PS Signalingdevelopment

per device

Android: from 10% 25% share

Great improvement in iOS & RIM signaling load … Android not so!Great improvement in iOS & RIM signaling load … Android not so!

- 35% signaling

+ 25% volume

Apple iOS

RIM

- 30% Signaling


PS Signaling … The network challenge?Remains a challenge for network aggregation points.

+140%

-50%+200%

CAGR +95% over period

IntroducingCELL-PCH 1

Introducing3GPP Fast Dormancy

1 NSN based feature.

Illustration

Much have been done on signaling … and “we” have gotten smarter.Much have been done on signaling … and “we” have gotten smarter.


3G Growth …will continue … for some timeand eventually decline as subs convert to LTE.3G growth …will continue

Illustration of a European Marketwith ca. 50+% prepaid base.

2006 2017

3GContract

3GPrepaid

CAGR 75%@ 2006 - 2011

Total 3G Data Traffic1

3G LTEConversion

CAGR 45% @ 2012 - 2017

2025

GSM 3G Conversion

1 Note: Due to the complex dynamics of technology migration and dependency on operator policy the phase-off of 3G is highly uncertain.

Illustration


Total growth … another leap with LTE.

Illustration of a European MarketLTE introduction 2013 earliest.

2012 2018 2025

Total Data Traffic

LTE CAGR 84%@ 2013 - 2018

CAGR 52% @ 2020 - 2025

3G LTE Conversionby 2025

500+ 2015 traffic@ 100% LTE share

LTE 2 3G Traffic@ 30% LTE share LTE

Illustration


10 20 40 60 85 120 120 120 120 120 120

0

5

10

15

2010 2012 2014 2016 2018 2020

Total spectrum in use for mobile data

Inc

rea

se

ove

r 20

10

Spectral Efficiency (*)

Spectral demand (limited)

Spectral demand (unlimited) Spectral demand could exceed spectral efficiency

between 2014 - 2016.

Spectral demand could exceed spectral efficiency

between 2014 - 2016.

When data demand exceeds spectral efficiency gains.”Houston we have problem”.

1 Mobile operator with (1) 20MHz @ 800MHz (LTE), (2) 20MHz @ 900MHz (2GHSPA),(3) 50MHz @ 1800MHz (2GLTE), (4) 30MHz @ 2100MHz (HSPA+). Total spectrum position 120 MHz.

Illustration of a European market 1

(*) realWireless report for Ofcom,: 4G Capacity Gains, Final Report, January 2011.

3G LTE LTE-aConversion

Leapfrog network capacity, e.g.,

Small cells topologies

Smart antennas

Early LTE deployment

Price, Control & Policy.

More spectrum.

Leapfrog network capacity, e.g.,

Small cells topologies

Smart antennas

Early LTE deployment

Price, Control & Policy.

More spectrum.

The spectrum crunch.

A lot moreComplexity, Capex and Opex

A lot moreComplexity, Capex and Opex

NOT GOODAT

ALL!


Data Mining – perception versus real experience.Tangible network factors impacting customer perception.

Satisfaction

Dis-satisfaction

Expectations

unfulfilled

Network Experience

Data

Behavioral

Data

Voice SMS

Data

Device

Financial

Data

CDRCSSR

Speed

Data QoE

Etc..

Customer

Service Data

Network State

From cell level up

Signaling Load

Mobility

Segmentation

Data


Expectations

fulfilled

Data Mining – customer perception versus experience.Strength of market survey data with hard network-centric data.

*Participants in the survey are informed and agreed (i.e., opt in policy applied) that their data will be used for research. No DPI applied.

Dissatisfied Groups Characteristics

< 90% of the time on 3G when using data.

Successful PDP context creations < 80%.

3G Voice Call Setup Duration > 3 seconds.

2G Voice Call Setup Duration > 5 seconds.

Postal code areas (i.e., coverage/capacity)

Handset type(e.g., iPhone 3GS and Blackberry 9700) .

Data usage > 300MB per month.

Number of sites visited > 60.

Voice call duration per month >450 minutes.

A relatively high bill. (i.e., higher bill, higher expectations)

Ca. 30% of active customer.

Re-prioritizing deployment.

Ca. 35+% of smartphones.

Dependency on perceived quality.

3G Coverage & Capacity.

< 5% of active customer.

Network Optimization.

Illustration


Out of Technology Scope

Data Mining – the Big Data picture 1.Capacity planning on the cell level using data mining strategies.

C1ell

C4ell

Cn-2ell

Illustration

n = 20,000 Cells 5 load-functions (output) 16 input cell-level parameters (input) Up-to 100,000 regression models. Planning validity < 4+ month

<Voice calls>, <R99 users>UL, DL

<HS-D/U-PA users>, Max HS-D/U-PA users,

Radio Resource Control Attempts*,

Radio Access Bearer (total, voice, data)

<Soft-HO area>, <DL / UL Speed>

<Voice / Data proportion originating in cell>

<Voice calls>, <R99 users>UL, DL

<HS-D/U-PA users>, Max HS-D/U-PA users,

Radio Resource Control Attempts*,

Radio Access Bearer (total, voice, data)

<Soft-HO area>, <DL / UL Speed>

<Voice / Data proportion originating in cell>

1. RAB release by interference2. Average Noise Raise (ANR)3. R99 specific ANR4. Consumed DL Power5. No Code Available

1. RAB release by interference2. Average Noise Raise (ANR)3. R99 specific ANR4. Consumed DL Power5. No Code Available

1 Paper on “Mass Scale Modeling for Prediction and Simulation of the Air-Interface Load in 3G Radio Access Networks”, by Radosavljevik, v.d. Putten & K. Kyllesbech Larsen submitted to The 18th ACM SIGKDD Conference on Knowledge Discovery and Data Mining’12, *One 1 RRC per active device.

Cell Input Xj (per hour).

Cell Output: Ci=1..5

𝑪𝒊=𝟏 ..𝟓❑ =∑

𝒋=𝟏

𝟏𝟔

𝒂𝒊𝒋 𝑿 𝒋 ∀𝒏

Cn-1ell

Cnell


The network state-equation .. is there such a thing?Calculating the critical driver limit for capacity demand.

Number of devices per cell.

Rate of concurrent instances of demand per unit time.

Number of devices per cell.

Rate of concurrent instances of demand per unit time.

Critical driver level (e.g., #devices) resulting in capacity upgrade is (simplified)

Critical driver level (e.g., #devices) resulting in capacity upgrade is (simplified)

* k is the number of standard deviation over the mean that is considered.

Illustration

Cell

(v1 ) Voice

(x1 ) R99

(y1 ) Signalling

(z1 ) H

SU

PA

(1 ) H

SD

PA

Effective rate pj

per device

Fundamental load drivers

Average load of cell j:

Solve for ncritical or pcritical

Average load of cell j:

Solve for ncritical or pcritical

nj #active devices

Ci Installed capacity


The network state-equation ...practical applications.

1 ….

Examples from a 3,000+ Node 3G network

The network (cell-based) state-equation allows reliable long-term 3G radio resource capacity planning.

The network (cell-based) state-equation allows reliable long-term 3G radio resource capacity planning.

0%

20%

40%

60%

80%

100%

0

50

100

150

200

250

300

350

Nod

e-Bs

Number of smart-phones per Node-B

Frequency

Cumulative %

Ca. 1,250 smart-phones per 3-carrier Node-B, carrier expansion should be expected.

2,500 smart-phones per 6-carrier Node-B, carrier expansion should be expected.

0%

20%

40%

60%

80%

100%

0

50

100

150

200

250

300

350

Nod

e-Bs

Number of smart-phones per Node-B

Frequency

Cumulative %

RRL limitfor Ultrasite

RRL limitfor Flex2

CE limit@ 128 CE

CE limit@ 256 CE

CE limit@ 396 CE


What about FUP? ...Fair? Effective? … or a FAD?

1 ….

Volume per User

FUP Limit

Throttling

Days of“normal” usage

FUP flavors

Hard volume-limit throttling.

BH throttling.

Service based (dpi) throttling.

Traffic de-prioritization … etc…

FUP flavors

Hard volume-limit throttling.

BH throttling.

Service based (dpi) throttling.

Traffic de-prioritization … etc…

Mobile FUP implementations might not be very efficientas a structural traffic management remedy.

Mobile FUP implementations might not be very efficientas a structural traffic management remedy.

Re-active remedy.

Typically capture <2% of users.

Does not address signaling challenge from smartphone Apps.

1 Mbps

64 kbps


Illustration

0 31

Subscription days per month

FUP or FAD?Volume-driven FUP (of today) has little structural impact.

1 Approx. log-norm distribute, 2 in max ¼ of the Top-20% cells.

2.5 Hour

1 Day

10 Days

100 Days

>2.5 Year

>25 Years

>250 Years

>2,500 Years

Days to reach 500MB

Days to reach 2GB

30 Days = Reset

98%< 2%

< 0.5%80% FUP Addressable

Daily usage per active user

Time toFUP limit

Illustration

Example:

2,000 FUP relevant users20,000 Cells in Network

50% of FUP in 20% of Cells1,000 FUP served by 4,000 Cells

BH mean value of users1 per cellis 185 in the Top 20% Cells.

1 FUP relevant customer would compete for resources with at least

185 others in the Busy Hour 2.


Pre- & post-FUP implementation.Marginal traffic reduction achieved with FUP.

00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Daily Tail Volume Profile>2 GB usage

Daily Active Customer Profile> 2 GB usage

-15% Drop

Max -10% Drop

Max -0.05% Drop

Post-FUP

Pre-FUP

Post-FUP

Pre-FUP

55% of total trafficMax. 0.25% of total

Active Base


Note: Fair Use Policy with throttling to 64kbps after limit has been reached.

Illustration

What we need to be passionate about.

Deep understanding of data traffic is crucial.Deep understanding of data traffic is crucial.

Customer usage, experience and imposed policies impact. Customer usage, experience and imposed policies impact.

Automation (data mining combined with machine learning) way forward.Automation (data mining combined with machine learning) way forward.

Ensuring best customer experience at all times & at lowest cost.Ensuring best customer experience at all times & at lowest cost.


Detecon is specialized in providing ICT management consulting services with the infrastructure of a global player.

Detecon Branch Offices

Detecon advises on the issues of strategy, organization, and technology design for Telecommunications and IT companies.

Established in 1977, Detecon is experienced, thanks to the successful realization of more than 6,000 projects.

Detecon is international, with worldwide representation, clients in 165 countries, and employees from more than 30 nations.

Detecon has in-depth knowledge of theindustry and a consulting approachoriented towards implementation and cooperation as partners.

Detecon is part ofDeutsche Telekom Group.

DETECON International GmbH

Contact: M.-A. Schultze Phone +49 160 8841957

[email protected]


mailto:[email protected]

Thank you for your interest!

Contact: [email protected]: +31 6 2409 5202http://nl.linkedin.com/in/kimklarsenhttp://www.slideshare.net/KimKyllesbechLarsen

Acknowledgement: I am indebted to Veli-Pekka Kroger and Dejan Radosavljevik for greatly enhancing this work with valuable discussions and sharp analytical insights. Last but not least I acknowledge my wife Eva Varadi for her great support and understanding during the creation of this presentation.

http://nl.linkedin.com/in/kimklarsen

http://nl.linkedin.com/in/kimklarsen

Capacity planning in mobile data networks experiencing exponential growth in demand

Technology

g devices

gtraffic illustration

g traffic distribution50

gtraffic volume0

gcells relative

kim kyllesbech larsen

mobile data traffic

data traffic trend