This talk is about “how we can exploit social information in content distribution systems”
description
Transcript of This talk is about “how we can exploit social information in content distribution systems”
Content Distribution based on Social Information
Rubén Cuevas, Eva Jaho, Carmen Guerrero and Ioannis StavrakakisUniversity Carlos III Madrid
National Kapodistrian University of Athens
Paris, 16th October 2008
This talk is about “how we can exploit social information in content distribution
systems”
Outline
Introduction SwarmTella OnMove Conclusions
Introduction
Does this really make sense? Success of Content Distribution Systems
P2P (Emule, BitTorrent), APLICATION LAYER MULTICAST, UGC Applications (YouTube)
Success of Social Applications Instant Messaging (MSN) Social Networks (FaceBook, LinkedIn,…)
People would use their social application for Content Distribution? Yes This makes sense No Give it up
How can we use social information in Content Distribution?
Identifying those users with similar social features General
Similar profession Similar hobbies Similar interests ….
Wireless Environments Similar Mobility Pattern
And exchange contents with them
Which benefits can we obtain?
Accuracy People Satisfaction Targeted Content Advertisement Retrieve contents that really fit my social
profile
Cooperation People collaborate if they get benefit from the
system
Resource Saving Avoiding flooding Avoiding downloading not desired contents
Our contribution
We will present two systems that exploit Social Information in Content Distribution
SWARMTELLA (UC3M) “Exploiting Social Information in P2P Content
Distribution”
ONMOVE (UC3M and NKUA) “Exploiting Social Information for Content
Distribution in Wireless Delay Tolerant Environments”
SwarmTella
SwarmTella General framework for distribution of
different type of content (file-sharing, VoD Distribution and Live Streaming)
Community scenarios It can be intended as a Recomendation
System Delivery techniques based on swarming Nodes initially organized in an
unstructured p2p Distributed mechanism for building
communities based on users common interests on contents: Ranking Algorithm
Ranking Algorithm
RA allows each node to identify other nodes with similar interest in a transparent way to the end user.
Each node generate a ranking of the other nodes.
Nodes with higher ranking means that have common interests to the local node.
It uses local information (light) Received search queries Swarm’s peers discovery
SPPiD Secure Permanent Peer-ID (SPPiD)
The public part of a Public/Private key pair.
Transparent to the end user, generated and just used by the application
This allows to keep connection with other nodes along different sessions
Long term robust structure of the communities, long life of the IDs.
Privacity Concerns Not Secure Permanent ID KAD User Ids Skype Mail Accounts MSN, FaceBook
Swarmtella Publication Mechanism
.swarmtella file with metadata of the available content.
The node with a new content generates the .swarmetella file and an ADVERTISEMENT message to be sent to the a limited number of nodes (highest ranked) in the community.
Swarmtella Searching Mechanism
Multiattribute semantic query to the highest ranked nodes in the community
If it fails, then flooding algorithm in unstructured p2p (gnutella like)
BW consumed
Query Hit Rate
Top peers and community members
content
SwarmTella
Next steps: Design Details (e.g. Swarm Partition) Real workload
Pattern of Encounters in Swarms Uptime Pattern of P2P nodes Plan Crawling BT Swarms
TUDarmstadt and UC3M
Swarmtella Implementation Validation in Controlled Environment
Emulab, ModelNet
OnMove
A novel protocol for content distribution in wireless delay-tolerant environments
It is designed for handheld devices mobile phones, PDAs, etc…
Multiple uses: Advertisement Platform UGC Distribution Entertainment On the Road
OnMove
DTN Scenario
Individual A may come in contact with individuals B, C and D in the morning for a duration of time t1.
Then she goes to the cinema and connects with other individuals for a duration of time t2.
In the evening she goes to the concert and meets other people for a duration of time t3.
t2
cinema
A B
GH
L
t3
concertA G
KMN
O
t1
university
C
B
D
A
DTN Scenario (cntd.)
A retrieves contents from B,C,D at the university
A stores them
A forwards the stored contents to B,G,H,L at the cinema
A forwards the stored contents to G,K,M,N,O at the concert
t2
cinema
A B
GH
L
t3
concertA G
KMN
O
t1
university
C
B
D
A
Social networks can be either studied as:
whole networks with all of the ties describing relations in a defined population, or as
egocentric networks describing the ties that one or more specific individuals have
OnMove is designed by considering egocentric or personal networks for each individual.
Social networking design of OnMove
Egocentric networks involve a focal individual (ego) and the individuals (alters) to which it is linked.
We study the exchange of data of the surrounding individual with the others in the group based on social interests.
Objectives: To increase speed of content dissemination To improve accuracy of content dissemination (align content
dissemination with users’ interests)
Egocentric Networks
iegocentric network of individual i
Content Exchange Procedure
When an individual comes in contact with other individuals in a social group (locality) She exchanges its social profile with the others. She has to decide from/to which node it is going to
download/upload contents.
The individual ranks the others individuals in the locality Download/Upload from/to highest ranked individual The ranking algorithm is the core of the content exchange
procedure, and should aim at increasing its effectiveness
C
B
D
AA
Ranking parameters in OnMove
Social Similarity (SS): Similarity of social details (profession, interests, hobbies) of individuals
Content Accuracy (CA): Alignment of contents received by an individual from other individuals to his/her interests
Pattern of Meetings (PM): Defined by the frequency and the duration of these encounters
Connection Quality (CQ): Available bandwidth, interferences, type of connection (e.g., WiFi, Bluetooth)
Ranking parameters in OnMove (cntd.)
Egocentric Betweenness Bi of individual i: Number of pairs of neighbors of i that are not directly connected to each other. Individuals with high value of egocentric betweenness have a lot
of influence in the network as a lot of other individuals depend on them to make connections with other people.
Average Egocentric Betweenness (B*):
t1
t2
t3
A
DC
B
H
LG
K
M
N
O
T
tii tB
TB
1
* )(1
Ranking neighbours in OnMove
Ranking metric for each individual: A weighted average of the previous parameters
Weights for each parameter are assigned differently in different application scenarios
**
)(
iBiCA
iCQiPMiSS
BwCAw
CQwPMwSSwiRank
Application scenarios
Advertisement Platform Objective: Maximize the dissemination of the
advertised content (photo, video, etc.) Relevant Parameters: B*, SS
File-Sharing on the Road: Objective: Find contents of interest to a node Relevant Parameters: SS, PM, CQ, CA
OnMove
Next steps: Configuration and optimization of the ranking
algorithm mechanism in several application scenarios.
Analyzing social profiles available on current systems such as FaceBook and exporting them to OnMove.
Evaluate OnMove in a real testbed. Crawdad data (e.g. Haggle Project)
Analysis of OnMove in multihop networks
Content Distribution based on Social Swarms
Rubén Cuevas, Eva Jaho, Rubén Cuevas, Eva Jaho, Carmen Guerrero and Ioannis StavrakakisUniversity Carlos III Madrid
National Kapodistrian University Athens
Paris, 16th October 2008