iscc2011 ioannis stavrakakis_ keynote

20
Exploiting Social Metrics Exploiting Social Metrics in Content Distribution Content Distribution Ioannis Stavrakakis Ioannis Stavrakakis National & Kapodistrian University of Athens Based on works with: Merkouris Karaliopoulos, Eva Jaho, Panagiotis Pandazopoulos, Pavlos Nikolopoulos, Therapon Papadimitriou June 30, 2011 1 ISCC Keynote Talk June 30, 2011 Corfu

Upload: achu59

Post on 28-Nov-2014

189 views

Category:

Documents


0 download

DESCRIPTION

 

TRANSCRIPT

Page 1: Iscc2011 ioannis stavrakakis_ keynote

Exploiting Social MetricsExploiting Social Metrics in

Content DistributionContent Distribution

Ioannis StavrakakisIoannis StavrakakisNational & Kapodistrian University of Athens

Based on works with: Merkouris Karaliopoulos, Eva Jaho, Panagiotis Pandazopoulos, Pavlos Nikolopoulos, Therapon Papadimitriou

June 30, 2011

1ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu

Page 2: Iscc2011 ioannis stavrakakis_ keynote

Focus on two key social metrics

Interest similarity

Centrality

2ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu

Page 3: Iscc2011 ioannis stavrakakis_ keynote

1 ‐ Interest Similarity Groups in online social networks formed to exchange information / content based on acquaintance, family relationships, social status, educational/professional backgroundbackground, ......... 

…yet interests/preferences of group members are not always similar

Is there value in assessing and exploiting interest similarity in groups?

3ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu

interest similarity in groups?

Page 4: Iscc2011 ioannis stavrakakis_ keynote

Define and measure Interest Similarityassess overall interest similarity in social groupsassess overall interest similarity in social groups

identify interest‐based structure within social groups

ISCoDE framework3

User Similarity

(commodity)community

ISCoDE framework3

User profiles

Similaritymetrics

ydetectionalgorithms

0

Thematic areas

1st step: user profiles weighted graph

3E. Jaho, M. Karaliopoulos, I. Stavrakakis. ISCoDe: a framework for interest similarity-based community detection in social networks Third International Workshop on Network Science for Communication Networks (INFOCOM

2nd step : weighted graph interest-similar groups

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 4

in social networks. Third International Workshop on Network Science for Communication Networks (INFOCOM-NetSciCom’11), Apr. 10-15, 2011, Shanghai.

Page 5: Iscc2011 ioannis stavrakakis_ keynote

Similarity metrics: PS vs. InvKL (Kullback Leibler)• Proportional Similarity (PS)

∑=M

jiji FFFFPS 11)(– PS : {Fi , Fj} [0,1]

• Inverse symmetrized KL divergence∑ ∑

= M Mm

jjm

ii

ji

FFFFInvKL 1),(

∑=

−−=m

mm FFFFPS12

1),(

y g– InvKL : {Fi, Fj} (0,∞) ∑ ∑

= =

+m m m

i

mm

j

mj

mm

i

FFF

FFF

1 1

loglog

mnF , 1≤n≤N, 1≤m≤M : distribution of node n over interest class m

Example with M=2 interest classes and N=2 nodes

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 55

• Proportional Similarity (PS) Inverse symmetrized KL divergence (InvKL)

Page 6: Iscc2011 ioannis stavrakakis_ keynote

Resolution performance

InvKL can identify smaller and more similar communities than PS, in a highly similar network, g y

PS can identify smaller communities than InvKL, in a highly dissimilar network

(could argue that this is not very useful)

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 66

useful)

Page 7: Iscc2011 ioannis stavrakakis_ keynote

Can Interest similarity improve network protocols ?

Gain of cooperation for content replication in a group of nodes1

T : tightness metric (=mean invKL), measuring interest i il it bsimilarity across group members

Percentage of cooperative nodes

1 E. Jaho, M. Karaliopoulos, I. Stavrakakis, “Social similarity as a driver for selfish, cooperative and

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 7

altruistic behavior”, in Proc. AOC 2010 (extended version submitted to IEEE TPDS)

Page 8: Iscc2011 ioannis stavrakakis_ keynote

Can Interest similarity improve network protocols ?

C di i i i i i k 2Content dissemination in opportunistic networks2

Protocols A,B,C are push protocols exercising interest-based forwarding

2S.M. Allen, M.J. Chorley, G.B. Colombo, E. Jaho, M. Karaliopoulos, I. Stavrakakis, R.M Whitaker, “Exploiting user

Protocols A,B,C are push protocols exercising interest based forwarding

(Precision: valuable received / total received --– Recall: valuable received / total potentially valuable generated)

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 8

interest similarity and social links for microblog forwarding in mobile opportunistic networks”, submitted to Elsevier PMC, 2011

Page 9: Iscc2011 ioannis stavrakakis_ keynote

2‐ Betweeness Centrality (BC)  

2.1 ‐ Content (service) Migration / Placement 

Can BC help provide for a low‐complexity, distributed, scalable solution?

Destination‐aware vs destination unaware BC

Ego‐centric vs socio‐centric computation of BC

9ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu

Page 10: Iscc2011 ioannis stavrakakis_ keynote

CBC: the “destination‐aware” counterpart to BC 

a measure of the importance of node's u social position : lies on

th li ki th

Betweenness Centrality (u ): portion of all pairs shortest paths of G that

paths linking others

pass through node u

ConditionalConditional Betweenness Centrality (u, t ) : portion of all shortest paths of

G from node u to target t, that pass through node u

10

a measure of the importance of node's u social position : ability to

control information flow towards

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 10

10target node

10

Page 11: Iscc2011 ioannis stavrakakis_ keynote

The content placement problem

Deploy  scalable and distributed mechanisms for publishing, placing, moving UG Service facilities / content within networking structures

Optimal content / service placement in a Graph  k‐median

Only distributed, scalable, solutions are relevant Use local information to migrate towards a better location

Use locally available limited information to solve repeatedly small‐scale k‐median and repeat

(*)K. Oikonomou, I. Stavrakakis, “Scalable Service Migration in Autonomic Network  Environments,” 

IEEE JSAC, Vol. 28, No. 1, Jan. 2010G S d ki N L t i K Oik I St k ki A B t “Di t ib t dG. Smaragdakis, N. Laoutaris, K. Oikonomou, I. Stavrakakis, A. Bestavros, “Distributed 

Server Migration for Scalable Internet Service Deployment”,  to appear in IEEE/ACM T‐ Net. (2011) , also in INFOCOM2007

11ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu

Page 12: Iscc2011 ioannis stavrakakis_ keynote

C id f d i h hi h CBC l

Centrality‐based service migration

Consider set of nodes with highest CBC values  

Solve iteratively small‐scale k‐medians

hosti є Gi

on subgraphs Gi Є G, around the 

current facility location of host i Gi

containing the top nodes based

on CBC values

Map the outside demand properly on nodes b h iin subgraphs Gi I

12P Pandazopoulos M Karaliopoulos I Stavrakakis “Centrality‐driven scalable service migration”

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 12

12

12

P. Pandazopoulos, M. Karaliopoulos, I. Stavrakakis, Centrality driven scalable service migration ,23rd International Teletraffic Congress (ITC), Sept. 6‐9, 2011, San Francisco, USA.

Page 13: Iscc2011 ioannis stavrakakis_ keynote

L h d f d i h!

simulation results: ISP topologies / non‐uniform load 

Less than a dozen of nodes is enough!

Demand load : Zipf distribution (with skewness s)Demand load : Zipf distribution (with skewness s)

Datasets correspond to different snapshots of 7 ISPs collected by mrinfo multicast tool *

* J.-J. Pansiot, P. Mérindol, B. Donnet, and O. Bonaventure, “Extracting intra-domain topology from mrinfo probing ” in Proc Passive and Active Measurement Conference (PAM) April 2010

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 13

probing, in Proc. Passive and Active Measurement Conference (PAM), April 2010.

13

Page 14: Iscc2011 ioannis stavrakakis_ keynote

Ego‐centric vs socio‐centric computation of BC

Very high rank correlation (Spearman coefficient) !!!

Ego‐ and socio‐ centric metrics identify same subsets

P. Pandazopoulos, M. Karaliopoulos, I. Stavrakakis, “Egocentric assessment of node centrality in physicalnetwork topologies” submitted to Globecom 2011

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 14

network topologies , submitted to Globecom 2011

Page 15: Iscc2011 ioannis stavrakakis_ keynote

Betweeness Centrality (BC)  2‐ Betweeness Centrality (BC)  

2.1 ‐ Centrality‐driven routing in opportunistic nets

(SimBetTS and BubbleRap use BC values of encounters for content forwarding)

How is performance of centrality‐based routing affected by

Adding or not, destination awareness to BC (BC vs CBC)

Working with ego‐centric vs socio‐centric BC values

Type of contact graph (unweighted vs weighted) ? Not discussed hereType of contact graph (unweighted vs. weighted) ? Not discussed here

P Nikolopoulos et al “How much off‐center are centrality metrics for opportunistic routing?”

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 15

P. Nikolopoulos, et.al.  How much off center are centrality metrics for opportunistic routing? , CHANTS 2011 Workshop (in MobiCom), Sept 23, 2011, Las Vegas

Page 16: Iscc2011 ioannis stavrakakis_ keynote

Datasets Datasets 

5 well‐known iMote‐based real traces available from the Haggle Project at CRAWDAD.

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 16

Page 17: Iscc2011 ioannis stavrakakis_ keynote

opt optimal routing through knowledge of contact sequences

BC BC vsvs CBCCBCopt  optimal routing through knowledge of contact sequences.

BC/CBC  up to 30% of messages never reach their destinationb t 5 ti h d 1 d f dditi l d labout 5 times more hops and 1 day of additional delay

BC outperforms CBC in delayCBC in delay (due to zero CBC values when destination in andestination in an unconnected cluster)

CBC outperforms BC in hops (up to 50% shorter50% shorter paths, due to selecting more proper nodes to 

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 17

p pforward to)

Page 18: Iscc2011 ioannis stavrakakis_ keynote

sociosocio‐‐ vsvs egoego‐‐metricsmetrics

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 18

Page 19: Iscc2011 ioannis stavrakakis_ keynote

sociosocio‐‐ vsvs egoego‐‐metricsmetrics

strong positive correlation of socio‐ and ego – metrics (Intel / Content data)

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 19

Page 20: Iscc2011 ioannis stavrakakis_ keynote

ConclusionsConclusionsFocused on exploring the impact on two key social metrics on content distribution

Interest SimilarityCentralityCentrality

Interest similarity metricsHighly similar groups can yield high gains in content replication.Interest similarity –based forwarding improves performanceWorth assessing interest similarity in groups – framework for doing that

D ti ti BCDestination‐aware BC :Very effective in content placement (BC is totally ineffective)Decreases hop count in opp nets (energy) substantially. Can increase delayp pp ( gy) y y

Ego‐centric centrality variants (BC/CBC)Highly rank correlated no performance degradation in content placement /centrality driven content forwardingcentrality‐driven content forwarding.Easier to compute

ISCC Keynote Talk ‐ June 30, 2011 ‐ Corfu 20