ComNets

Dimensioning of the Shared Transport Network for Collocated Multiradio: LTE and HSDPA

Ming Li, M.Sc. FFV Workshop 2012-II

October 19th, Hamburg

X. Li, M. Li, U. Toseef, D. Dulas, M. Nowacki, A. Timm-Giel, C. Goerg, R. Ruchala, “Dimensioning of the Shared Transport

Network for Collocated Multiradio: LTE and HSDPA” in the WiMob 2012, Barcelona, October, 2012.

ComNets

Shared Transport Network for Collocated Multiradio

2

Different radio technologies (2G, 3G, LTE)

will still coexist for some time

How to aggregate the traffic?

How much gain do we have?

How to dimension the common link?

Shared Transport Networks(IP Based)

Internet Cloud

We propose a multiradio shared transport system

• Reduced costs on renting transport resources

• Maximized reuse of existing infrastructure

• Provide smooth migration to new

technologies (LTE and LTE-Advanced)

ComNets

Overview of our Simulation Model

3

ComNets

Our Focuses and Contributions

• Our Focuses – Focus on the elastic traffic (TCP-based traffic)

– Cover the User-Plane interfaces (LTE: S1; HSPA: Iub)

– Only the DL direction is considered

• Our Contributions – Investigate the shared transport network

• Proposed a QoS framework

• Analysis of the achievable sharing gain (bandwidth saving)

– Propose analytical models to dimension the shared link • Reduce the dimensioning time from hours/days to several seconds

• Portable dimensioning tools are implemented

4

ComNets

Process Sharing Model for Elastic Traffic

• The dimensioning model for elastic traffic (TCP-based traffic) is based on M/G/R-PS Model

– Due to TCP flow control all TCP flows share the link bandwidth equally the system can be modeled as a Processor Sharing system

– The M/G/R-PS Model has been used for dimensioning

• networks, e.g. IP network, ADSL, etc.

• mobile networks, e.g. UMTS

M/G/R-PS Model

Processor Sharing-Poisson Arrivals

Departures

1

2

R

Peak data rate

File length

Link utilization Delay factor

Number of servers

R = C / rk

Average File Transfer Delay of QoS class k

2 ( , )( ) 1

(1 )

k k k k kk k

k k k k

x E R R xT x f

r R r

Xi Li (PhD thesis): “Radio Access Network Dimensioning for 3G UMTS”. Vieweg+Teubner Verlag, 2011.

5

R = C / rk

C: Link capacity

rk : Flow peak data rate

ComNets

WFQ Scheduler

AF31 PHB

AF21 PHB

AF11 PHB

BE PHB

Transport Network

IP DiffServ QoS Class based on PHBs

(Per Hop Behavior)

Transport Scheduler

Dimensioning Models

6

Weighted

M/G/1-PS

M/G/R-PS

Extended M/G/R-PS

for DiffServ with WFQ

HSPA

LTE

ComNets

Dimensioning Framework for Elastic Traffic (nGBR)

LTE Air Interface

Model

Weighted

M/G/1-PS

Shared Transport

Model

Extended M/G/R-PS

for DiffServ with WFQ 𝑟𝐼𝑢𝑏_𝐻𝑆

Traffic load per

PHB

WFQ weight

File size

Shared transport capacity

App. delay per PHB TN

Delay

Model

TN delay per PHB • Cell capacity

• QCI types

• Traffic model

• QCI weights

• …

HSDPA Air Interface

Model

M/G/R-PS

• Cell capacity

• Cell load

• UE category

• UE peak data

rate

• #UEs

scheduled

per TTI

• …

𝑟𝑆1_𝐿𝑇𝐸(𝑘)

eNB/NB

RNC/

aGW

Shared TN

7

ComNets

Modeling the LTE air interface

Estimate the avg. UE Uu throughput of Elastic traffic (nGBR) with Weighted M/G/1-PS Model

o Model Parameters for elastic traffic

- Air interface capacity of a LTE cell: CUu_LTE

- the defined QCI weight for the QoS class k in LTE MAC scheduler: WQCI(k)

- Mean traffic load of the elastic traffic of QoS class k: LUu_LTE(k)

a) Step 1: calculate the available cell capacity used for elastic traffic of QCI class k

b) Step 2: calculate the estimated peak data rate at the LTE S1 interface for each QCI class k

8

_ _ _ _

( )( ) max , ( )

( )

QCI

Uu LTE Uu LTE Uu LTE Uu LTE

j kQCI

i

w kC k C C L j

w i

1_ _ _( ) ( ) ( )S LTE Uu LTE Uu LTEr k C k L k

ComNets

Estimate the avg. UE Uu throughput of Elastic traffic (nGBR) with M/G/R-PS Model

o Model Parameters for elastic traffic

- Air interface capacity of a HSDPA cell: CUu_HS

- Peak data rate (bits/s) of a HSDPA bearer (TCP flow): rHS

- Mean traffic load of all elastic traffic in a cell: LUu_HS

a) Step 1: calculate the air interface delay factor

b) Step 2: calculate the estimated peak data rate on the Iub interface

Modeling the HSDPA air interface

HSUu

HSUu

HS

HS

HSUu

HS

HSHS

HSHSHS

HSUu

C

Land

r

CRwhere

R

RREf

_

__

2

_)1(

),(1

HSUuHSHSIub frr __ /

1

0

2

!!

!),(R

i

Ri

R

AR

R

R

A

i

A

AR

R

R

A

ARE

# Servers Link Utilization Delay factor

Erlang’s second formula (Erlang C formula)

9

ComNets

Modeling of the Shared Transport Network

The dimensioning model for elastic traffic (TCP-based traffic) is based on extended M/G/R-PS Model*

( ) ( ) / ( )TN TN TNj L j BW j

( )( ) max , ( )

( )

WFQ

TN TN TN TN

i jWFQ

i

w jBW j BW BW L i

w i

2 ( , ( ))( ) 1 ( )

( ) (1 ( )) ( )

j j j TN j

j TN

TN j TN TN

x E R R j xT x f j

r j R j r j

( ) / ( )j TN TNR BW j r j

* Extended M/G/R-PS K. Lindberger (1999)

File Size # Servers

Peak data rate Link Utilization Delay factor

app. mean

delay for PHB

class j

(sojourn time)

10

rS1_LTE(k) or rIub_HS

ComNets

Validation of the Analytical Dimensioning Models

11

LTE Part, 1GB TCP buffer size, 20MHz cell bandwidth

Services Load per cell PHB (WFQ weight)

FTP (QCI9) 8, 12, 16Mbps BE (1)

HSDPA Part, 64KB TCP buffer size, 5MHz cell bandwidth

Services Load per cell PHB (WFQ weight)

FTP 6Mbps AF21 (100)

HSPA only LTE only Shared TN Model0

20

40

60

80

100

120

Transport Network Bandwidth [Mbps]

Scenarios

Ba

nd

wid

th [M

bp

s]

8Mbps 12Mbps 16Mbps0

20

40

60

80

100

120

Transport Network Bandwidth [Mbps]

LTE cell load [Mbps]

Ba

nd

wid

th [M

bp

s]

Analytical

Simulation

Analytical

Simulation

* Dimensioning target: HSDPA 50ms, LTE 15ms 44% 55% 66% 77%

0

5

10

15

TN

Dela

y [

ms]

BE PHB: Analytical

BE PHB: Simulation

AF21 PHB: Analytical

AF21 PHB: Simulation

44% 55% 66% 77%10

12

14

16

18

20

Transport link utilizationF

ile d

ow

nlo

ad

dela

y [

s]

BE PHB: Analytical

BE PHB: Simulation

AF21 PHB: Analytical

AF21 PHB: Simulation

ComNets

Analysis of the Achievable Sharing Gain

12

0

5

10

15

20

25

30

35

40

1 2 3 4 5 6 7 8

Sh

ari

ng

ga

in [

%]

LTE/HSDPA cell load [Mbps]

LTE/HSDPA Sharing Gain

0

2

4

6

8

10

12

14

16

18

20

10 20 30 50 70 80 90

Sh

ari

ng

ga

in [

%]

HSDPA load share [%]

LTE/HSDPA Sharing Gain

Sharing gain decreases with the increased cell loads.

Higher when HSDPA percentage is higher than LTE

highest when equal load

* Dimensioning target: HSDPA 15ms, LTE 15ms

𝑆ℎ𝑎𝑟𝑖𝑛𝑔_𝐺 = 1 −𝐵𝑊𝑆ℎ𝑎𝑟𝑒𝑑_𝑇𝑁

𝐵𝑊𝐻𝑆 + 𝐵𝑊𝐿𝑇𝐸⋅ 100%

ComNets

Summary

• A multiradio shared transport system is presented, and developed in a system simulation model in OPNET.

– Both application and Transport network delays are improved

– Around 20~30% gain on bandwidth saving from the example scenarios

• Propose an analytical framework to dimension the shared transport interface for the elastic traffic based on the extended M/G/R-PS model

– Modeling the LTE/HSDPA radio interface

– Modeling the DiffServ QoS scheme in shared transport network

– Dimension the shared transport network to fulfill the QoE/QoS requirements

• The presented validation results demonstrate the proposed analytical models

– Can appropriately estimate the application/link level performances of different traffic and priorities

– Can estimate the achievable LTE/HSDPA sharing gain from sharing the transport

13

ComNets

Next Steps

• Extend the dimensioning models for multiple base stations (eNBs, NBs) scenarios, and other traffic types

• Adding other radio technologies like HSUPA, GSM and UMTS Rel’99

14

ComNets

Thanks and Any Questions?

15