Dynamics of Social-aware Pervasive Networks

http://copelabs.ulusofona.pt

Human-centered Computing Lab

Dynamics of Social-aware Pervasive Networks

Waldir Moreira and Paulo [email protected]

March 27th, 20154th IEEE PerCom Workshop on the Impact of Human Mobility in Pervasive Systems and Applications (PerMoby 2015)St. Louis, USA

Waldir Moreira, [email protected] http://copelabs.ulusofona.pt

Agenda

Introduction

Human Behavior-aware Aggregation

HBA Analysis

Impact of HBA on Opportunistic Forwading

Conclusions and Future Work


Introduction

Opp. forwarding influenced by user mobility and social engagement

Aggregation algorithms identify relevant nodes and links

If oblivious about dynamism

– Nodes with same degree

– Links represent a mere encounter

When dynamism is considered

– Nodes with different degrees

– Social relevant links


Introduction

Complex Network Analysis used to study graph properties

– Opp. forwarding performs well over small-world/scale-free graphs

Previous work study the dynamics of underlying graphs, but

– Social level is a mere product of age and frequency of contacts

– Social aggregation based on graph density (finding optimal density)

– Looks at the global network behavior (lower granularity)

However, user behavior is rather intricate

– Users interact differently throughout their daily routines


Introduction

Social aggregation upon such a dynamic behavior

– Social level has more to it (age/frequency of contacts vs. time spent socially engaging)

– Patterns in user behavior exist and are influential

– Varying users’ social interactions and their respective mobility are of paramount importance

Thus, we focus on the time-evolving property of user social behavior

– Human behavior-aware aggregation

– Mobility patterns in different moments in time

– To understand the impact on the operation of opp. forwarding


Human Behavior-aware Aggregation

Takes into account

– Daily sample: time interval in which users have similar behavior

– Social intensity: duration of contacts between nodes

– Time-evolving social ties: variation of users’ behavior


HBA Analysis

Done with the Gephi v0.8.21 and Cytoscape v2.8.32 tools

Over two CRAWDAD human traces

– Cambridge, which comprises a group of 36 students carrying iMote devices during a two-month period in Cambridge, UK

– MIT, which comprises 97 Nokia 6600 smart phones distributed among the students and staff of this institution

Goal

– Identify the type of complex network formed with HBA in different time periods of a user daily routine


HBA Analysis

Time-evolving property is imperative (Cambridge traces)

– If overlooked, homogeneity trend in terms of node degree

– If considered, highlights only the socially well-connected nodes


HBA Analysis

Complex network features (Small-world features)

– SW properties vary with HBA but are still there

– Nodes are less clustered and avg. path length is higher

– Lower, and yet relevant, number of links


Parameters Values

Simulator Opportunistic Network Environment (ONE)

Routing Proposals HBA-based fwd, dLife, dLifeComm, Bubble Rap, Rank and Epidemic

Scenarios CRAWDAD Cambridge trace CRAWDAD MIT trace

Simulation Time 1036800 sec 17020800 sec

# of Nodes 36 97

Generated messages 6000 78000

Node Interface Bluetooth

Node Buffer 2 MB

Message TTL Length of experiments

Message Size 1 – 100 kB


HBA-based Forwarding (HF): weight of social link dLife: social weight, node importance dLifeComm: social weight, node importance, net. community Bubble Rap: network community, node centrality Rank: node importance Epidemic: pure contact-based forwarding

Understand the impact on opp. forwarding when operating overHBA-based graphs



Average latency reduction of 5.1% for all forwarding approaches

Average cost

– Reduction of 44% for approaches based on social weights or communities (HF, dLife, Bubble Rap)

– Social-oblivious and node degree-based approaches are highly penalized (68% increase)

• Epidemic, ignores dynamics and spreading is eased by nodes being well conected (giant component, SF)

• dLifeComm and Rank, significant number of nodes with degree higher than the average

Average delivery probability

– Subtle decrease: up to 3.6% and 4.3% for the Cambridge and MIT

– Data mule effect for approaches relying on social weights (HF, dLife, and dLifeComm): carriers have difficulty in finding nodes with a higher social weight → direct delivery


Conclusions and Future Work

More coarseness in the analysis of the dynamics of user social behavior

– Social aggregation-based graphs must take this into account

– Social-oblivious and node degree-based approaches are punished with significant cost

HBA does have a positive effect on the performance of opp. forwarding

– Reflects the dynamics of pervasive networks, presenting small-world properties with edges reflecting high social intensity

– Occasional data muling for social weight-based approaches

Future work

– HBA with other forwarding schemes and other sets of human traces

– Compare the intensity-based aggregation with density-based aggregation: social weights vs. age/or frequency of contact

Dynamics of Social-aware Pervasive Networks

Technology

user behavior

social ties

dynamic behavior social

nodes time

variation of users behavior

user mobility

impact of human mobility

time interval