Information-Centric Networking (ICN) has gained increasing attention from the research community as it is able to improve content dissemination by releasing the dependency on content location. With the current host-based Internet architecture, networking faces limitations in dynamic scenarios, due mostly to host mobility. The ICN paradigm mitigates such problems by releasing the need to have an end-to-end transport session established during the life time of the data transfer. Moreover, the ICN concept solves the mismatch between the Internet architecture and the way users would like to use it: currently a user needs to know the topological location of the hosts involved in the communication when he/she just wants to get the data, independently of its location. Most of the research efforts aim to come up with a stable ICN architecture in fixed networks, with few examples in ad-hoc and vehicular networks. However, the Internet is becoming more pervasive with powerful personal mobile devices that allow users to form dynamic networks in which content may be exchanged at all times and with low cost. Such pervasive wireless networks suffer with different levels of disruption given user mobility, physical obstacles, lack of cooperation, intermittent connectivity, among others. This paper discusses the combination of content knowledge (e.g., type and interested parties) and social awareness within opportunistic networking as to drive the deployment of ICN solutions in disruptive networking scenarios. With this goal in mind, we go over few examples of social-aware content-based opportunistic networking proposals that consider social awareness to allow content dissemination independently of the level of network disruption. To show how much content knowledge can improve social-based solutions, we illustrate by means of simulation some content-oblivious/oriented proposals in scenarios based on synthetic mobility patterns and real human traces. This presentation was given in the 8th IEEE WoWMoM Workshop on Autonomic and Opportunistic Communications (AOC 2014), on June 16th, 2014 in Sydney, Australia.

1. http://copelabs.ulusofona.pt
Human-centered Computing Lab
Social-aware Forwarding in Opportunistic Wireless
Networks: Content Awareness or Obliviousness?
Waldir Moreira and Paulo Mendes
waldir.junior@ulusofona.pt
June 16th, 2014
8th IEEE WoWMoM Workshop on Autonomic and Opportunistic Communications (AOC 2014)
Sydney, Australia

2. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt
Agenda

Introduction

Social-aware Forwarding in
Opportunistic Wireless Networks

Experiments

Content Awareness or Obliviousness:
Which way to go?

3. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt
Introduction

Information-Centric Networking (ICN): driven by information
- Data-Oriented Network Architecture (DONA)
- Named Data Networking (NDN)
- Network of Information (NetInf)

Opportunistic Networks (OppNets): driven by contact opportunity
- Single-copy forwading
- Replication-based
Aim at improving content dissemination

4. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt
Introduction

Powerful wireless devices

Wireless technology for direct
communication
- Vehicle-to-vehicle, Wi-Fi direct,
device-to-device

Dynamic scenario: content
prosumers (producers and
consumers)
- Content exchanged anytime and
anywhere, highly mobile users,
intermittent connectivity, physical
obstacles, …

5. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt
Introduction

Motivation: by looking at the nature and properties
- OppNets → contact opportunities, cope with disruptive networking
scenarios, persistent storage
- ICN → content knowledge, availability, security, and location-
independence, in-network caching
- Both → abstract the need for e2e communication
- Alignment with today's scenario

Goal: bring together the best of both worlds and understand how
much content knowledge can impact the performance of social-
based opportunistic forwarding

6. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt
Social-aware Forwarding in
Opportunistic Wireless Networks

Different flavors
- Epidemic solutions
- Based on contact history
- Social similarity

Our focus: social-aware family
- Avoid the volatile property of mobility → dynamic scenarios
- Common social groups and communities, node popularity, levels of
centrality, shared interests, and future social interactions

In what concerns content knowledge
- Content-oblivious: completely unaware of content information
» Bubble Rap, CiPRO, and dLife
- Content-oriented: consider different levels of content knowledge
» SocialCast, ContentPlace, and SCORP

7. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt

Assess the impact of combining content and social awareness to
forward data in opportunistic networks
Parameters Values
Simulator Opportunistic Network Environment (ONE)
Routing Proposals Bubble Rap, dLife and SCORP
Scenarios CRAWDAD Cambridge trace Synthetic mobility
Simulation Time 1036800 sec 345600 sec
# of Nodes 36 150
Mobility models Human Shortest Path Map Based
Generated messages 35, 175, 350, 700, and 1225 200
Node Buffer 2 MB
Node Interface Bluetooth
Message Size 1 – 100 kB
Message TTL Length of experiments
K-Clique, k 5 (Bubble Rap)
K-Clique, familiarThreshold 700 sec (Bubble Rap)
Daily samples 24 (dLife and SCORP)
Experiments

8. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt

Bubble Rap
- 1msg/int: Comm.-based fwd
- Buffer use: 9.94MB (5msg/int)
→ 50% reduction

dLife
- 1msg/int:
Social weight, importance
→ direct delivery
- Buffer use: 2.17MB (10msg/int) → 43% reduction

SCORP
- 1msg/int: dense network → Interest-based forwarding
- Buffer use: 0.16MB (35msg/int) → stable delivery
Experiments
Impact of network load

9. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt

1msg/int: comm/direct delivery
→ Low number of replicas

Bubble Rap and dLife
- Cost is expected to increase
as load increases
- Peaks due to message creation
time and contact sporadicity
- High number of contacts → more replications

SCORP
- More interest → better message carrier
- Avg. of 6.39 replicas, across all msg/int configurations, against 452.41
and 96 replicas of Bubble Rap and dLife
Experiments
Impact of network load

10. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt

1msg/int: few contacts and
long periods (12 to 23 hours) of
almost no contact
→ high latency

As load increases
→ messages created in
different time periods

Bubble Rap and dLife
- Latency in function of delivered messages plus choice of forwarders
→ decrease and variable behavior

SCORP
- More interest → better forwarder → higher probability of coming into
contact with other nodes sharing similar → reducing latency
Experiments
Impact of network load

11. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt

Bubble Rap
- Del. based on global centrality
- Few nodes plus constrained
buffer and long TTL
→ reduced delivery

dLife
- Nodes move (i.e., interact) less
→ longer time to converge
→ more replicas → buffer exhaustion
- Social awareness is advantageous, but not enough

SCORP
- Social weights with others sharing interests → no impact from mobility

100000-second pause time: few and sporadic interactions (intervals of
20 and 26 hours) → convergence time
Experiments
Impact of node mobility

12. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt

Bubble Rap and dLife
- As pause time increases,
the # of contacts decreases
→ Better view of the network
in termos of social metrics
→ less replicas

SCORP
- Interest-based replication + socially well-connected nodes
→ very low replication (average of 0.5 replicas)
- Resource use: estimated average occupancy of 0.03MB/node/day

100000-second pause time: cost is in function of delivered messages
- Low deliveries due to contact sporadicity → proposals have a low cost.
Experiments
Impact of node mobility

13. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt

Bubble Rap, dLife, SCORP
- Less interactions among nodes
→ longer time to converge
→ messages are stored longer
→ increased latency experience

100000-second pause time:
latency peaks are due to contacts happening
in a sporadic fashion with intervals between them of up to 26 hours
Experiments
Impact of node mobility

14. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt

Definitely, content knowledge (i.e., type and interested parties)
must be taken into account

The content-oriented social-aware opportunistic forwarding scheme
experiences delivery improvements up to 60%, while its latency and
cost can be reduced by 75% and 90% respectively, when compared
to content-oblivious forwarding schemes

Results do show the advantages of building social- and content-
aware forwarding schemes for networking in disruptive scenarios
Content Awareness or Obliviousness:
Which way to go?

15. Waldir Moreira, waldir.junior@ulusofona.pt http://copelabs.ulusofona.pt
Acknowledgements
To CitySense project from COPELABS and
Fundação de Amparo à Pesquisa do Estado de
Goiás (FAPEG)