PRECISE: Power Aware Dynamic Traffic Steering inTightly Coupled LTE Wi-Fi Networks
Thomas Valerrian Pasca S, Himank Gupta, Bheemarjuna ReddyTamma and Antony Franklin A
Department of Computer Science and EngineeringIndian Institute of Technology - Hyderabad
PIMRC - 2017 PRECISE IITH 1 / 27
Outline
1 Introduction
2 Challenges in C-LWIP Architecture
3 Proposed SolutionIM PhaseGI PhasePRECISE
4 Performance Evaluation
5 Conclusions
PIMRC - 2017 PRECISE IITH 2 / 27
Introduction
Introduction
Mobile data traffic growth is exploding and it will reach 35Exabytes per month by 2020 [1]Telco providers/operators face challenges in order to improve theirnetwork capacitiesUtilizing unlicensed band efficiently has gained operator interestfor the increasing their bandwidth
Offloading traffic from cellular network toWi-Fi network has become operator sweetspot for handling the demandWe focus on LWIP for harvesting the benefitsof unlicensed band
PIMRC - 2017 PRECISE IITH 3 / 27
Introduction
Multi RAT Aggregation
Application Layer
Protocol Stack Optimization
Intelligence is required to send the traffic
over multiple interfaces based on metrics
such as RTT and bandwidth
Medium access techniques can be enhanced,
when one link regulates and coordinates
medium access of the other link
Application should be intelligent to open
multiple sockets on different IP addresses
and reorder the data at application layer
Traffic offload algorithms based on
network state can be implemented for
effective utilization of multiple links
Replace your text here! Replace your text here! Replace your
text here!
Replace your text here! Replace your text here! Replace your
text here!
Your Text
Application layeraggregation (eg.,Samsung boost)Transport layeraggregation (eg.,MPTCP)IP layer aggregationloosely coupled (eg.,PMIP, ANDSF)IP layer aggregationtightly coupled (eg.,LWIP)
PIMRC - 2017 PRECISE IITH 4 / 27
Introduction
LTE Wi-Fi interworking
Internet EPC
WLAN eNB
UE
NSWO S2a S1 S2a/S2b S1 S1NSWO NSWO
Tunneling
R12 eSaMOG
(CN Based)
CN Based
NB-IFOM & ANDSF Enhancement
RAN Based
Integrated 3GPP-WLAN RATs
1
1Ref: Intel presentation at TSDSIPIMRC - 2017 PRECISE IITH 5 / 27
Introduction
LWIP Architecture
LWIP has following benefitsWi-Fi operations are controlled directlyvia LTE base station (eNB) andtherefore LTE core network (i.e.,Evolved Packet Core (EPC)) need notmanage Wi-Fi separatelyRadio level integration allows effectiveradio resource management acrossWi-Fi and LTE linksLTE acts as the licensed-anchor pointfor any UE, providing unified connectionmanagement with the network
LWIP has finer level of control on radiointerfaces, for making efficient steeringdecision
S1 U
Figure : C-LWIPArchitecture
PIMRC - 2017 PRECISE IITH 6 / 27
Challenges in C-LWIP Architecture
Challenges in C-LWIP Architecture
Co-tier interference due to densification of small cellsQoS provisioning for users in high interference zoneIndependent deployment of LTE and Wi-Fi nodes results in poorinterference mitigationSteering the traffic across LTE and Wi-Fi alone does not suffice.
PIMRC - 2017 PRECISE IITH 7 / 27
Proposed Solution
Proposed Solution - PRECISE
Two Phase of Optimization i.e. IM Phase and GI PhaseIM Phase: Sets the Optimal transmit Power for LTE and Wi-Fiinterfaces to avoid co-tier interferenceGI Phase: Sets the transmit power of LTE and Wi-Fi interface inorder to meet the GBR requirementsSteering: Efficient traffic steering to aggregate the link benefits
IM Phase GI Phase
LTE Coverage in
IM Phase
Wi-Fi Coverage
in IM Phase
Coverage expanded
by GI Phase
Coverage reduced
by GI Phase
A
CB
D
PIMRC - 2017 PRECISE IITH 8 / 27
Proposed Solution IM Phase
Interference Mitigation (IM) Phase I
Maximize the SINR for the users by varying the power of LTE andWi-Fi links
A fractional frequency reuse pattern of LTE and Wi-Fi coveragemaximizes the users SINR
Maximize ΘIM =
U,B∑i=1,j=1
(αLi,j × SINRL
i + αWi,j × SINRW
i ) (1)
PIMRC - 2017 PRECISE IITH 9 / 27
Proposed Solution IM Phase
Interference Mitigation (IM) Phase II
s.t.B∑
j=1
αLi,j ≤ 1 ∀i and
B∑j=1
αWi,j ≤ 1 ∀i (2)
αLi,j =
{1, if SINRL
i ≥ ThLTE
0, otherwise(3)
αWi,j =
{1, if SINRW
i ≥ ThWi−Fi
0, otherwise(4)
αLi,j =
{0 or 1, if αW
i,j = 10, otherwise
(5)
αWi,j =
{0 or 1, if αL
i,j = 10, otherwise
(6)
PLmin ≤ PL
j ≤ PLmax ; PW
min ≤ PWj ≤ PW
max (7)PIMRC - 2017 PRECISE IITH 10 / 27
Proposed Solution GI Phase
GBR Improvement (GI) Phase
Objective function maximizes the weighted SINR function whereweight is proportional to number of unsatisfied GBR
Maximize ΘGI =
U,B∑i=1,j=1
(rLi,j × SINRL
i + rWi,j × SINRW
i ) (8)
s.t. {SINRL
i − (γ ×Θ(SINRLi )) ≥ 0 if Θ(SINRL
i ) ≥ SM
SINRLi −Θ(SINRL
i ) ≥ 0 otherwise(9)
{SINRW
i − γ ×Θ(SINRWi ) ≥ 0 if Θ(SINRW
i ) ≥ SM
SINRWi −Θ(SINRW
i ) ≥ 0 otherwise(10)
PLmin ≤ PL
j ≤ PLmax ; PW
min ≤ PWj ≤ PL
max (11)
PIMRC - 2017 PRECISE IITH 11 / 27
Proposed Solution GI Phase
Traffic Steering : TOPSIS
Input: Set of all flow (Fi ) parameters, Link to which flow affinityhas to be obtained
1: Vector Normalization of all flow parameters Fi,j where i ∈ {flows1,. . . ,k}, j ∈ {network parameters}
2: Apply given set of weights wT = {w1, . . . ,wn}3: Fi,j ← Fi,j × wj4: Find A+ (Positive ideal solution) and A− (Negative ideal solution)5: Find Positive ideal separation (S+) and Negative Ideal separation
(S−)6: Calculate Ci for each flow: Ci ← Si−
Si++Si−7: AI ← sort {Ci } in descending order8: Return the flow affinity index AI i for every flow Fi
PIMRC - 2017 PRECISE IITH 12 / 27
Proposed Solution PRECISE
PRECISE
Trigger IM
Phase
Start
Obtain all Flow Parameters
Gs >=90%
Trigger GI
Phase
TOPSIS
Load LTE >
Load Wi-Fi
Steer unmet GBR
flows to Wi-Fi
Is GBR unmet?
Steer set of NGBR
flows with high AI
to Wi-Fi
Is GBR unmet?
Steer set of NGBR
flows with high AI
to LTE
Steer unmet GBR
flows to LTE
Stop
After
experiment
duration
PIMRC - 2017 PRECISE IITH 13 / 27
Proposed Solution PRECISE
Experimental Setup
Parameter Value
# of UEs, LWIP Nodes 100, 4Max Tx power of LTE & Wi-Fi 23, 20 dBm
LTE path loss model 3GPP indoor path loss modelWi-Fi path loss model ITU path loss modelLTE MAC Scheduler Priority Set Scheduler (PSS)
UE position RandomUE mobility model Constant Position Mobility Model
Wi-Fi Standard IEEE 802.11nWi-Fi frequency and bandwidth 2.4 GHz, 20 MHzLTE frequency and bandwidth 2.6 GHz, 10 MHz
PIMRC - 2017 PRECISE IITH 14 / 27
Proposed Solution PRECISE
Simulation Setup
0
50
40
300
1020
20
3010
40
0 50
5
10
-40
-20
0
20
40
60
X: 26
Y: 25
Z: 7
X: 35
Y: 41
Z: 1
X: 42
Y: 17
Z: 7
X: 11
Y: 23
Z: 4
Figure : Building dimensionconsidered for evaluation
Building of dimension 50m X50m X 10mBuilding has two floors and awall per every 10m4 C-LWIP nodes placed withmean euclidean distancebetween nodes as 20mPath loss model includes walland floor losses
PIMRC - 2017 PRECISE IITH 15 / 27
Performance Evaluation
SINR Distribution
0
0.2
0.4
0.6
0.8
1
-20 -10 0 10 20 30 40 50
CD
F
SINR (dB)
Fixed Power (LTE)PRECISE (LTE)
Fixed Power (Wi-Fi)PRECISE (Wi-Fi)
Figure : CDF of UE SINR.
SINR distribution observed infollowing 2 Cases
LWIP with fixed powerLWIP with optimal powerobtained from PRECISEalgorithm
PRECISE has improved SINRof UEs by 4 dB in both LTEand Wi-Fi linksThis improvement in SINR isdue to optimal power control
PIMRC - 2017 PRECISE IITH 16 / 27
Performance Evaluation
Ensure GBR in the network
0
20
40
60
80
100
120
140
# o
f G
BR
flo
ws n
ot m
et
GI Phase TriggerLoad=100Load=300Load=600
0
20
40
60
80
100
120
140
0 2 4 6 8 10
Thro
ughput (M
bps)
Time (Seconds)
Load = 100Load = 300Load = 600
Figure : Events triggered for varyingload.
When no of flows are 100, IMphase is triggered morefrequently since there are lessunsatisfied GBR flows.As number of flows increases,more GI Phase is triggeredmore which regulates thetransmit power inorder toreduce unsatisfied GBR Flows.
PIMRC - 2017 PRECISE IITH 17 / 27
Performance Evaluation
Different Phases of PRECISE Algorithm
0
10
20
30
40
50
100 200 300 400 500 600
# o
f T
rig
gers
Load (# of flows in the network)
IM-Phase TriggersGI-Phase Triggers
Figure : Triggers for differentthreshold.
In every 200 msec either of IMPhase or GI Phase will betriggeredAs number of flows increasesfrom 100 to 600, the numberof IM Phase trigger reducessince more unsatisfied GBRflows exists.PRECISE instantiates more GItriggers to reduce unmet GBRflows
PIMRC - 2017 PRECISE IITH 18 / 27
Performance Evaluation
Throughput Analysis
0
0.2
0.4
0.6
0.8
1
0 0.5 1 1.5 2 2.5 3 3.5
CD
F
User Throughput (Mbps)
Wi-Fi PreferredRel-12
α-OptimalPRECISE
Figure : CDF of UEs throughput.
CDF of UEs throughputobserved for a fixed number ofFlows i.e 600.Rel-12 ensures that UE isassociated with best interfaceand flows are routed through itα-Optimal only steer the flowsacross interface proportionallyfair based on link ratesPRECISE regulates thetransmit power and alsoenables flow steering acrossbest interface
PIMRC - 2017 PRECISE IITH 19 / 27
Performance Evaluation
Throughput Analysis (contd.)
0
20
40
60
80
100
120
140
160
300 350 400 450 500 550 600
Thro
ughput (M
bps)
Load (# of flows in the network)
Wi-Fi PreferredRel-12
α-OptimalPRECISE
Figure : Throughput for different load.
0
50
100
150
200
300 350 400 450 500 550 600N
um
ber
of U
nsatisfied G
BR
Load (# of flows in the network)
Wi-Fi PreferredRel-12
α-OptimalPRECISE
Figure : Unsatisfied GBR.
PIMRC - 2017 PRECISE IITH 20 / 27
Performance Evaluation
Throughput and GBR Analysis
Performances of different algorithms are compared with PRECISEalgorithm for 600 Flows.PRECISE algorithm has improved the network throughput by 48%and 84 % as compared to α-optimal Algorithm and 3GPP REL-12respectively.PRECISE algorithm minimizes the unmet GBR flows compared toother algorithms beacause of PRECISE alogrithm regulates thetransmit power of LWIP node and steer the flows to best availableinterface. While α-optimal only steer the flows and Rel-12 choosethe interface beased on best SINR available.PRECISE algorithm has reduced the number of GBR flows unmetby 35% as compared to α-optimal scheduler.
PIMRC - 2017 PRECISE IITH 21 / 27
Conclusions
Conclusions and Future Directions
Co-located LWIP enables sophisticated control over LTE andWi-Fi RATs.PRECISE algorithm thrives to ensure QoS and maximizesthroughput in co-located LWIP deployment scenario.PRECISE algorithm employs power control for interferencemitigation and to enhance GBR throughput of UEs with poorSINR.PRECISE supports dynamic flow steering using MADM techniquein order to improve the network throughput.PRECISE outperformed throughput of α-optimal scheduler by48% and 3GPP Rel-12 interworking by 84%.PRECISE reduced the number of unmet GBRs by 35% comparedto α-optimal scheduler.PRECISE can be optimized further to improve UE battery savings.
PIMRC - 2017 PRECISE IITH 22 / 27
Conclusions
Acknowledgements
This work was supported by the project "Converged CloudCommunication Technologies"
PIMRC - 2017 PRECISE IITH 23 / 27
Conclusions
References I
[1] Cisco. (2017) Visual Networking Index: Global Mobile Data TrafficForecast Update. [Online]. Available: https://goo.gl/zY4nKI[2] 3GPP, "LTE-WLAN Aggregation and RAN Controlled LTE-WLANInterworking," Tech. Rep. 36.300, 2016.[3] S. Thomas et al., "Architectural Challenges and Solutions forCollocated LWIP - A Network Layer Perspective," in 2017 Twenty ThirdNational Conference on Communication (NCC), March 2017, pp. 1-6.[4] S. Deb et al., "Algorithms for eICIC in LTE HetNets," IEEE/ACMTrans. on Net., vol. 22, no. 1, pp. 137-150, 2014.[5] J.-H. Lee et al., "Host-based distributed mobility management:Example of traffic offloading," in IEEE CCNC, Jan 2013, pp. 637-640.[6] K. Lee et al., "Mobile Data Offloading: How Much Can Wi-FiDeliver?" IEEE/ACM Trans. on Net., vol. 21, no. 2, pp. 536-550, April2013.
PIMRC - 2017 PRECISE IITH 24 / 27
Conclusions
References II
[7] S. Singh et al., "Optimal traffic aggregation in multi-RATheterogeneous wireless networks," in IEEE, ICC, May 2016, pp.626-631.[8] S. Singh et al., "Proportional Fair Traffic Splitting and Aggregation inHeterogeneous Wireless Networks," IEEE Communications Letters,vol. 20, no. 5, pp. 1010-1013, May 2016.[9] Y.-J. Lai, T.-Y. Liu, and C.-L. Hwang, "Topsis for MADM," EuropeanJournal of Operational Research, vol. 76, no. 3, pp. 486-500, 1994.[10] 5GAmericas. (2015, Nov) Whitepaper on LTE Aggregation andUnlicensed spectrum. [Online]. Available: http://goo.gl/4yHPhH[11] 3GPP, "LTE/WLAN Radio Interworking" Tech. Rep. 37.834, 2013.[12] Skype. (2017) How much bandwidth does Skype need?[Online].Available: https://goo.gl/KYCZ5V
PIMRC - 2017 PRECISE IITH 25 / 27
Conclusions
PIMRC - 2017 PRECISE IITH 26 / 27
Conclusions
For Further Queries Contact US
www.thomasvalerrianpasca.in
https://github.com/ThomasValerrianPasca/
PIMRC - 2017 PRECISE IITH 27 / 27