Social-aware Utility-based Radio Resource Management
description
Transcript of Social-aware Utility-based Radio Resource Management
12 Jul. 2011 THALES - SoComNets 1
Social-aware Utility-based Radio Social-aware Utility-based Radio Resource ManagementResource Management
Timotheos Kastrinogiannis
School of Electrical and Computer Engineering National Technical University of Athens (NTUA)
12 Jul. 2011 2THALES - SoComNets
At a glance…
Content Management Networks User’s social behavior; Content attributes, content management goals and new users’
satisfaction criteria.
Physical Networks (i.e., underlay networking paradigms) Network-centric resource allocation and QoS provisioning mechanisms; Network Utility Maximization (NUM) theory.
* Top-down Cross Layering….* Top-down Cross Layering….
12 Jul. 2011 3THALES - SoComNets
Network Centric RRM (The Traditional View)
Rate
Control
Channel Estimation
Connectivity Management
Topology Control
Capacity Regions
Modulation & Coding
Schemes
Power Control
Spectrum Management
Users Assignment
Handover Coordination
Stochastic Control / Scheduling
Quality of Service & Fairness Management
Optimization Theory
Convex Optimization
Network Utility
Maximization
Linear Programming
Game Theory
Layering As
Optimization Decomposition
The Need….. The Evolution…..
12 Jul. 2011 4THALES - SoComNets
A Mathematical Theory of Network Architectures…
The first unifying view and systematic approach Network: Generalized NUM Layering architecture: Decomposition scheme Layers: Decomposed subproblems Interfaces: Functions of primal or dual variables
(KellyMaulloTan98):
12 Jul. 2011 5THALES - SoComNets
Designing a Utility ….
A utility function is based on three key components :
Indicators correlated to data transmission (π.χ. BER).
Indicators correlated to QoS prerequisites fulfillment Indicators correlated to preemption (e.g. real time aplications)
* * * *1 , 1 , ,( ,..., ,..., ,..., ) ( ,..., ) ( ,..., ) ( ,..., )i l m n TL i l SL i l m PE i m nU x x x x U x x U x x U x x
12 Jul. 2011 6THALES - SoComNets
Fundamental Utility Functions
Non-real-time Services
Concave
SigmoidalConvex
Real-time Services
12 Jul. 2011 7THALES - SoComNets
Methodology… (how to break the barrier?)
Content Management Networks
Social Networks (OSN)
Social-Aware Utility Functions
Integrated Heterogeneous Wireless Networks
Radio Resource Management
Overlay user-centric & content-centric network
Underlay user-driven communication network
Social Graphs Attributes
Content-Aware UtilityQoE-Aware Utility
Functions
Content Attributes Behaviourall Attributes
A novel Top-down Cross Layering
e.g., U( Service QoS, Centrality Metrics)
e.g., U( Service QoS, Content Attribute)
U( Service QoS, QoE)
12 Jul. 2011 8THALES - SoComNets
User’s Behavioral Attributes in Content Management (Case I)
Motivation: “Minimizing Node Churn in Peer-to-Peer Streaming”[1]
In P2P networks suffer from highly transient peers who join and leave the system (churn) at their own will.
In P2P streaming services content management should be performed in short time scales.
C. Vassilakis, I. Stavrakakis, “Minimizing Node Churn in Peer-to-Peer Streaming Computer Communications”, Computer Communications Journal, Vol.33, Iss. 14, pp.
1598-1614, Sept. 2010.
Node ChurnAffects System Performance
Is Affected by Node’s Service Quality
Dealing withthe effects of node
churn
Protocols Design
Peers Selection Strategy
Dealing withits causes
12 Jul. 2011 9THALES - SoComNets
Thread of Analysis
Peers Selection Strategy
Node Churn Probability
MOSOff-line function
f (total number of dropped (lost) and repeated (frozen) frames)
Short-term On-line Functionf (Average Achieved Goodput, Target Rates, QoS Requirement Fulfilment)
Utility-Based
Radio Resource Managament
Implicit Calculation
Explicit Calculation
Short-term Goodput QoE Utility (video)
Churn Probability
Can we improve Node’s Churn Probability via RRM ?
12 Jul. 2011 10THALES - SoComNets
Corresponding Optimization*
,1
max1
max
max
max ( , , , )
. .
0 1,2,...,
0 1,2,...,
N
i i i iR P
i
N
ii
i
i i
U R P a b
s t P P
P P i N
R R i N
1, , min, , ,
,
,
( ) ( ) if ( ) ( 1)ˆ ( )
iWRT i i i RT i i RT i RT i
RT i
RT i
b A NodeChurn t b b t B t Tr Bb t
b otherwise
QoE-aware (Content Network-aware) RRM
Goal: Minimize Node’s Churn Probability
* *
, , , ,ˆ ˆˆ ˆ( , , ( )) ( , ( )) i iR R
RT i RT i RT i RT i RTU P a b t U P b t i S
QoS-Aware RRM
11
, ,1
( ) ( ) i
i
WWRT i RT i RT
k
B t t k i S
12 Jul. 2011 11THALES - SoComNets
Networking Results
Correlation with overlay network’s benefits…..
System Av. Goodput (per cell) (kbps)
0
100
200
300
400
500
600
700
800
30 Users 35 Users 40 Users
RRM
RRM_SDUA
System Av. Throughput (per Cell) (kbps)
0
100
200
300
400
500
600
700
800
30 Users 35 Users 40 Users
RRM
RRM_SDUA
System Churn Rate (departures/sec)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
30 Users 35 Users 40 Users
RRM
RRM_SDUA
12 Jul. 2011 12THALES - SoComNets
Beginning with a Simple Question….
“By what criteria do we evaluate a particular network architecture?”
Shenker
1995
VS
“The internet was designed to meet the needs of users, and so any evaluative criteria must reduce to the following question: how happy does this architecture make the users?”.
12 Jul. 2011 13THALES - SoComNets
The Evolutionary Path of QoS Initially,Initially,
correlation of services or service classes (application layer) with several network metrics (e.g. threshold-based values for latency, jitter, packet loss, e.t.c)
design of dynamic resource allocation algorithms that aim at maintaining these metrics at acceptable levels.
Then,Then, utility functions define a formal mathematical vehicle towards expressing and quantifying
user’s degree of satisfaction with respect to their multi-criteria service performance.
the goal of network design can be restated as being, quite simply, to maximize the sum of utilities, leading to the establishment of
Network Utility Maximization (NUM)Network Utility Maximization (NUM)
“ “a concrete theoretic framework”a concrete theoretic framework”
12 Jul. 2011 14THALES - SoComNets
Quality of Experience (QoE) (Case II)
Finally,Finally, Quality of Experience (QoE), defined as “a measure of the overall acceptability
of an application or service, as perceived subjectively by the end-user” ITU
How can we correlate QoS and QoE?
“E2E QoSE2E QoS” (enabled via monitoring and proper network reacting mechanisms);
QoE is enabled via the mappingmapping of users’ opinions for the quality of a service (in a normalized way e.g. MOSMOS)) to specific networking metrics (leading to proactive approaches);
Dynamic adaptationDynamic adaptation of a network’s operation and performance, in line with users’ dynamic requests (e.g. adaptive video resolution on demand).
12 Jul. 2011 15THALES - SoComNets
In Reality … Experience is Experience is SubjectiveSubjective & & Context-dependentContext-dependent
Cultural BackgroundCultural Background
Socioeconomic StatusSocioeconomic Status
Personal Experiences Personal Experiences Mood Mood
BackgroundBackground NoiseNoise
Multimedia Multimedia ContentContent
……..
Social DistanceSocial Distance Social BehaviorSocial BehaviorContentContent
ManagementManagement
12 Jul. 2011 16THALES - SoComNets
A Practical Idea
We envision the role of QoE as the vehicle that interconnects users/humans, applications and QoS-aware resource management mechanisms.
We propose a QoE framework that allows users to dynamically and dynamically and asynchronously express their (dis)satisfactionasynchronously express their (dis)satisfaction with respect to the instantaneous experience of their service quality at the overall network QoS-aware resource allocation process.
12 Jul. 2011 17THALES - SoComNets
Design & User InterfaceDynamic Utility Adaptation
NUM – Optimization
Feasibility&
Policies
Resource Allocation Adaptation
Pricing
Inc (+0.03 $ / min)
Dec (-0.02 $ / min)
Current Performance Level
Max Performance Level
Stepwise Increasing Performance& Corresponding Service Cost ($/min) Inc.
Feasibility of the Request
12 Jul. 2011 18THALES - SoComNets
QoE-aware Resource Allocation
Users’ 16, 17, 18 Achieved Goodput Cells’ Overall Goodput
12 Jul. 2011 19THALES - SoComNets
Towards the Second Direction
Content Management Networks
Social Networks (OSN)
Social-Aware Utility Functions
Integrated Heterogeneous Wireless Networks
Radio Resource Management
Overlay user-centric & content-centric network
Underlay user-driven communication network
Social Graphs Attributes
Content-Aware UtilityQoE-Aware Utility
Functions
Content Attributes Behaviourall Attributes
A novel Top-down Cross Layering
e.g., U( Service QoS, Centrality Metrics)
e.g., U( Service QoS, Content Attribute)
U( Service QoS, QoE)
12 Jul. 2011 20THALES - SoComNets
Social-aware Utility-based RRM (Case III)
Motivation: “Scalable Distribution of Content Updates over a Mobile Social Networks” Service: “Dynamic Content Distribution” Co-operation: subscribers to this service share their updates in an
opportunistic fashion Problem: “how the service provider can allocate its bandwidth
optimally to make the content at users as “fresh” as possible.”
12 Jul. 2011 21THALES - SoComNets
System’s Behavior (Under Optimal Content Management)
μ : total injection rate allocated among different users
Users are indexed according to their contact rates, in decreasing order.
Observations: For low values of μ the “most social” user accumulates all the injected rate, thus acting as a global hub of all incoming information;
Under certain conditions, it is actually optimal to allocate no bandwidth to the most social users in the system.
Idea: Social-aware Utility Functions that expressA. User’s QoS Prerequisites B. User’s Priority on Available Radio Resources with Respect to his Role/Importance in the Mechanism Of the Underlay Content Management Network.
Social Users’ Normalized Priority Indicator:
Non-Social Users’ Normalized Priority Indicator:
2
( )
2,1 ( ) ( / sec)
( )
log( 1)
ci i Thr
i
e if CR t CR contactsI t
otherwise
12 Jul. 2011 22THALES - SoComNets
Problem Setting… (initial experimentation…)
where
, , , min,ˆ ( ) ( ) ( )RT i RT i i i RT i ib t b A I t b b t
*
,1
max1
max
max
max ( , , , )
. .
0 1,2,...,
0 1,2,...,
N
i i i iR P
i
N
ii
i
i i
U R P a b
s t P P
P P i N
R R i N
* *
, , , ,ˆ ˆˆ ˆ( , , ( )) ( , ( )) i iR R
RT i RT i RT i RT i RTU P a b t U P b t i S
2
( )
2,1 ( ) ( / sec)
( )
log( 1)
ci i Thr
i
e if CR t CR contactsI t
otherwise
12 Jul. 2011 23THALES - SoComNets
Thank you…Thank you…