ikom - ComNets
Treffen/Workshop der ITG Fachgruppe 5.2.1
Radio Access Network Dimensioning for 3G UMTS Xi Li
[email protected]
November 13, 2009
ikom - ComNets
Outline
Introduction and Motivation UMTS Network Dimensioning Framework Developed Simulation Models Developed Analytical Models Dimensioning Models and Results Conclusions and Outlook
2
ikom - ComNets
Outline
Introduction and Motivation UMTS Network Dimensioning Framework Developed Simulation Models Developed Analytical Models Dimensioning Models and Results Conclusions and Outlook
3
Universal Mobile Telecommunication System (UMTS) ikom - ComNets UE UE
Node B
Circuit Switched Domain
Iub
PSTN ...
UE
RNC UE
UE
Node B
Iub
Packet Switched Domain
Internet X.25 ...
UE UE
UE UE Node B RNC
UTRAN User Equipment Base Station Radio Network Controller
Core Network UTRAN PSTN
External Networks
UMTS Terrestrial Radio Access Network Public Switched Telephone Network
4
Motivation of UMTS Network Dimensioning ikom - ComNets
Dimensioning: determine appropriate bandwidths for transport links maximizing utilization of transport resources guarantee QoS (Quality of Service) requirements The transport resource within the UTRAN is considerably costly
UTRAN
Iub Interface
Costly Costlyinterface interface Strict Strictdelay delayQoS QoS
Dimensioning of Iub is important to design a high cost- efficient UMTS network
5
Goal of This Thesis ikom - ComNets
UMTS network is developing fast
Evolutions of UMTS
Radio Access Network (RAN) evolution: Rel99, HSDPA, HSUPA, HSPA+, LTE Evolved UMTS terminals and emerging new services Significant increase of the traffic volume Remarkable changes in traffic pattern and characteristics
Transport Technologies for UTRAN, e.g. migration from ATM to IP Quality of Service Schemes, e.g. QoS differentiation and prioritization
Goal Goalof ofthis thisThesis Thesis
Investigate Investigateimportant importantaspects aspectsrelated relatedtotothe theIub Iubdimensioning dimensioning Develop Developdimensioning dimensioningapproaches approachesfor fordifferent differentUMTS UMTSNetworks Networks simulation models simulation models analytical analyticalmodels models Derive Deriveimportant importantdimensioning dimensioningguidelines guidelinesand andrules rules
6
ikom - ComNets
Outline
Introduction and Motivation UMTS Network Dimensioning Framework Developed Simulation Models Developed Analytical Models Dimensioning Models and Results Conclusions and Outlook
7
Objectives of UMTS Network Dimensioning ikom - ComNets
Network Costs: the costs correlated with the expenditures necessary for leasing transport link bandwidths
Quality of Service
user-relevant QoS: refers to the QoS related to the individual users
Application delay or throughput, connection reject ratio due to admission control function
network-relevant QoS: network-specific QoS to evaluate the quality of a network, measured on the packet level
Packet delay, packet loss ratio
The The goal goal of of network network dimensioning dimensioning is is to to minimize minimize costs costs while while maximizing maximizing QoS QoS
8
Framework of UMTS Network Dimensioning ikom - ComNets Simulation Approach
Analytical Approach
Input
Output
Traffic Demand traffic class traffic load traffic distribution
Dimensioning Process
QoS Targets
Network Cost minimum required link capacities (Mbit/s)
user-relevant QoS network-relevant QoS
Network Configurations
QoS
network topology traffic control functions resource control functions transport technology
Bandwidth
QoS mechanisms 9
ikom - ComNets
Outline
Introduction and Motivation UMTS Network Dimensioning Framework Developed Simulation Models Developed Analytical Models Dimensioning Models and Results Conclusions and Outlook
10
Simulation Models
ikom - ComNets
Model a complete UMTS system following 3GPP specifications Focused on a detailed modeling of the Iub interface (i.e. protocol stack, transport network, resource and QoS management) Modeling of air interface and core network are simplified Reduce complexity and improve simulation efficiency
IP Transport
ATM Transport
ATM Transport
IP Transport 11
ikom - ComNets
Outline
Introduction and Motivation UMTS Network Dimensioning Framework Developed Simulation Models Developed Analytical Models Dimensioning Models and Results Conclusions and Outlook
12
Services and QoS Measures ikom - ComNets Voice
Video Conferencing
Web
FTP
Real Time (RT)
Non Real Time (NRT)
low delay low loss require Admission Control
carried by TCP/IP delay tolerant
Circuit-Switched Traffic
Applications/Services
Traffic Classes
Elastic Traffic
Blocking probability (CAC reject ratio)
Application Throughput (Application Delay)
QoS Measures at flow/call level
Userrelevant QoS
Packet Delay Packet Loss ratio
Packet Delay Packet Loss ratio
QoS Measures at packet level over the Iub
Networkrelevant QoS
13
Overview of Analytical Models ikom - ComNets
Dimensioning Tool
Analytical Models
User-Relevant QoS
Traffic Scenario
Circuit-Switched Traffic
QoS Measure
blocking QoS
Proposed Analytical Models
Elastic Traffic
application delay or throughput
Erlang Loss Model
Processor Sharing (PS) Model
Erlang-B MD Erlang-B
M/G/R-PS queuing model
Network-Relevant QoS
Mixed Traffic
both QoS need to be met
Processor Sharing Model + Erlang Model
Traffic Policy - BW sharing - BW separation
Modeling Call or Flow Level
Circuitswitched traffic
Elastic Traffic
Mixed Traffic
packet delay, packet loss ratio over the Iub interface
Queuing Models with nonMarkovian Arrival Process
Non-preemptive priority queuing model
MMPP(2)/D/1
MMPP(2)/D/1 - Priority
or BMAP/D/1
or BMAP/D/1-Priority
Modeling Packet Level 14
ikom - ComNets
Outline
Introduction and Motivation UMTS Network Dimensioning Framework Developed Simulation Models Developed Analytical Models Dimensioning Models and Results
Processor Sharing Model (Application Performance) Packet level Queuing Model (Transport Network Performance)
Conclusions and Outlook
15
Processor Sharing (PS) Model for Elastic Traffic - for User-Relevant QoS (Application Performance) ikom - ComNets UE
Radio Access Bearer (RAB)
rpeak
rpeak UE
Assumptions
rpeak
R = C / rpeak
rpeak UE
Iub (C) NodeB
rpeak UE
RNC
Flow arrival follows Poisson Process General file length distribution
Radio Network Cont
rpeak
Expected Sojourn Time (average transfer delay) UE
M/G/R-PS Model File length
E M / G / R {T ( x)} =
Peak data rate
Number of servers R = C / r peak
x ⎛ E 2 ( R, R ρ ) ⎞ x ⎜⎜1 + ⎟⎟ = ⋅ f R r peak ⎝ R (1 − ρ ) ⎠ r peak
Link utilization
K. Lindberger (1999)
Delay factor 16
Proposed Extensions on M/G/R-PS Model ikom - ComNets Seven Extensions are proposed in this thesis to incorporate UMTS networks Case
Extensions
Analytical Realizations RTTadjust = RTT ⋅ f R
1. Single RAB No CAC
New parameter UL_rtt_ratio E {T ( x )}* = E {T ( x )} + ( 2 + UL _ rtt _ ratio ) RTT adjust
2. Single RAB With CAC 3. Multiple RABs No CAC
General M/G/R-PS model - R is bearer specific - consider total traffic
4. Rate Adaptation - BRA
Reuse single rate M/G/R-PS Calculate an average rate from different rpeak to derive R RAB CAC
Radio Access Bearer Call Admission Control
Ri =
C
∑ ρ
ρ =
r i
E M / G / R {T ( xi )} =
r peak _ avg =
i
bearers
xi ⎛ E 2 ( Ri , Ri ρ ) ⎞ xi ⎜1 + ⎟ = f Ri r i ⎜⎝ Ri (1 − ρ ) ⎠⎟ r i
K
∑ r ⋅ q j
j =1
j
Ravg = C / r peak _ avg
17
Proposed Extensions on M/G/R-PS Model (cont’) ikom - ComNets Case 5. Mixing with CS Traffic
Extensions
Analytical Realizations M/G/R-PS
(a) C Iub = C elastic + LCS (b) C Iub
= C elastic + C CS M/G/R-PS
Erlang
Last mile links
Node B
C ac_1
Node B
C ac_2
6. Multi-Iub RAN
Backbone Link C bb IP Router
RNC
C ac_n
Node B
7. IP DiffServ
E M / G / R {T ( xk )} =
⎛ E 2 ( Rk , Rk ρ k ) ⎞ xk ⎜1 + ⎟= f k r peak _ k ⎜⎝ Rk (1 − ρ k ) ⎠⎟ r peak _ k xk
18
IP-based UTRAN with DiffServ QoS Structure ikom - ComNets
RNC
Node B
UMTS Core Network
Per Hop Behavior (PHB) EF AF PHB
Expedited Forwarding Assured Forwarding Per Hop Behavior
SP WFQ DiffServ
Strict Priority Weighted Fair Queuing Differentiated Services
19
Validation of Application Delay Estimation ikom - ComNets AF21 PHB - NRT RAB 128kbps
AF11 PHB - NRT RAB 64kbps
Single Link Scenario
12
12 ) 10 s ( y a 8 l e d . p 6 p a 1 4 1 F A
M/G/R/N-PS Simulations
) s (
y a l e d . p p a 1 2 F A
2
0 0.5
WFQ weight
10
M/G/R/N-PS Simulations
8 6 4 2
0.6
0.7 0.8 Iub link utilization
0.9
0 0.5
1
0.6
AF41 PHB - NRT RAB 384kbps
Service class
PHB
RT voice
EF
10 )
RT video
EF
8
NRT RAB 64kbps
AF11
20
NRT RAB 128kbps
AF21
30
NRT RAB 256kbps
AF31
40
NRT RAB 384kbps
AF41
50
NRT HSPA 2Mbps
BE
10
1
0.9
1
12
M/G/R/N-PS Simulations
10 ) s ( y a l e d . p p a E B
6 4
M/G/R/N-PS Simulations
8 6 4 2
2 0 0.5
0.9
BE PHB HSPA
12 s ( y a l e d . p p a 1 4 F A
0.7 0.8 Iub link utilization
0.6
0.7 0.8 Iub link utilization
0.9
1
0 0.5
0.6
0.7 0.8 Iub link utilization
The Therelative relativeerrors errorsof ofobtained obtainedanalytical analyticalresults resultsare arewithin within the theagreed agreedlevel levelfor fornetwork networkdimensioning dimensioningof ofindustry industry
20
Queuing Models for Network-Relevant QoS
ikom - ComNets
FP PDUs
Delay distribution
TTI
DCH 1
TTI
DCH 2
0.99 Link
DCH n
TTI
AAL2 ATM Queue Segmentation Queue Deterministic Deterministic service rate service rate
30ms
Arrivals
Departure Depatures
RT or NRT
H
Packet scheduling: Non-preemptive priority
Departure Depatures
Arrivals Server process (deterministic service rate)
(a) Single-service system
RT
queuing delay
NRT L
Server process (deterministic service rate)
(b) Priority system
Arrival process model (shall capture bursty and self-similarity of the aggregated arrival traffic and Bulk Arrival of packets) 2-state Markov Modulated Poisson Process (MMPP) model, where the interarrival time distribution is based on 2-Phase Hyper-exponential distribution Batch Markovian Arrival Process (BMAP) 21
MMPP Model for Estimation of the Iub delay ikom - ComNets Capture of the Characteristic of the Arrival Traffic
Traffic demand
Add network / protocol overhead
Mean traffic
Variance
Correlation
Measure arrival traffic
Capture the arrival traffic characteristics MMPP arrival process model parameters
MMPPD/1 queuing MMPP/D/1 – priority queuing
Queuing delay distribution
22
Validation of the Iub Delay Estimation ikom - ComNets Scenario I: 100% voice traffic (single Iub) Traffic model: Adaptive Multi Rate (AMR) 12.2kbps Speech/silence period: exponential distribution, mean = 3 seconds Call duration: exponential distribution, mean = 120 seconds Dimension QoS target: 99% of packets experience less than 10ms Iub delay voice only scenario - Rel99 ATM-based Iub 10000 ] s p b k [ h t d i w d n a b b u I d e r i u q e r
9000 8000 7000
l e d 0.35 o m l a 0.3 c i t y l a 0.25 n a e 0.2 h t f o r 0.15 o r r e 0.1 e v i t a 0.05 l e r
system simulation M/D/1 H2/D/1 MMPP/D/1
6000 5000 4000 3000 2000 1000 0 0
1000 2000 3000 voice traffic demand [kbps]
voice only scenario - Rel99 ATM-based I ub 0.4
4000
0 0
M/D/1 H2/D/1 MMPP/D/1
1000 2000 3000 voice traffic demand [kbps]
4000
23
Validation of the Iub Delay Estimation ikom - ComNets Dimension QoS target: 99% of voice packets experience less than 10ms Iub delay 99% of data packets experience less than 30ms Iub delay Scenario III: 90% web traffic (low priority) & 10% voice traffic (high priority)
Scenario II: 100% web traffic packet switched traffic (BRA) only
packet switched traffic (BRA) with 10% voice
8000 ] s 7000 p b k [ 6000 h t d i w d 5000 n a b b 4000 u I
8000
System simulation Queueing simulation (Opnet) Analytical calculation
] 7000 s p b k [ 6000 h t d i 5000 w d n a 4000 b b u 3000 I
d e r 3000 i u q e 2000 r
1000 1000
System simulation Queueing simulation (Opnet) Analytical calculation
d e r i u 2000 q e r
1000
2000 3000 4000 UTRAN traffic demand [kbps]
5000
0 0
1000 2000 3000 4000 UTRAN traffic demand [kbps]
5000
24
ikom - ComNets
Outline
Introduction and Motivation UMTS Network Dimensioning Framework Developed Simulation Models Developed Analytical Models Dimensioning Models and Results Conclusions and Outlook
25
Conclusions and Outlook
ikom - ComNets
Propose a general UMTS network dimensioning framework Develop several detailed simulation models with OPNET Investigate important aspects related to Iub dimensioning Develop novel analytical models for dimensioning of the Iub interface and validated by simulations
Circuit-switched Traffic User QoS
Network QoS
Elastic traffic
Mixed traffic
Blocking QoS Erlang-B (single stream) MD-Erlang B (multi-stream)
Application QoS Processor Sharing (PS) Extensions of M/G/R-PS
Consider applied traffic policy and its Mux. gain
Packet delay & loss QoS
Packet delay & loss QoS
MMPP: Markov Modulated Poisson Process BMAP: Batch Markovian Arrival Process
MMPP/D/1 or BMAP/D/1
MMPP/D/1 or BMAP/D/1
MMPP/D/1-nonpreemptive priority
Develop a dimensioning tool (in Matlab)
Summarize all proposed analytical models Apply to derive dimensioning rules 26
Conclusions and Outlook (cont’) ikom - ComNets
Dimensioning and Comparing of ATM- and IP-based UTRAN
Single Iub link scenario
Multi-Iub RAN scenario
Dimensioning HSPA traffic in ATM-based UTRAN
HSDPA HSUPA HSPA+Rel99 (Traffic Separation)
Further Work: Long Term Evolution (LTE)
Expect a much higher demand on transport bandwidth in access networks Dimensioning for LTE transport access network Investigating applicability of current dimensioning models Extensions of analytical models are desired
27
ikom - ComNets
Thank for your Attention
28
