social-aware opportunistic routing protocol based on user's interactions and interests
DESCRIPTION
Nowadays, routing proposals must deal with a panoply of heterogeneous devices, intermittent connectivity, and the users' constant need for communication, even in rather challenging networking scenarios. Thus, we propose a Social-aware Content-based Opportunistic Routing Protocol, SCORP, that considers the users' social interaction and their interests to improve data delivery in urban, dense scenarios. Through simulations, using synthetic mobility and human traces scenarios, we compare the performance of our solution against other two social-aware solutions, dLife and Bubble Rap, and the social-oblivious Spray and Wait, in order to show that the combination of social awareness and content knowledge can be bene cial when disseminating data in challenging networks This presentation was given on my behalf by Dr. Mendes in the 5th International Conference on Ad Hoc Networks (ADHOCNETS 2013), on October 16th, 2013 in Barcelona, Spain.TRANSCRIPT
Waldir Moreira and Paulo Mendes [email protected] [email protected]
5th International Conference on Ad Hoc Networks
October 16th, 2013
Social-aware Opportunistic Routing Protocol based on User's Interactions and Interests
2
Motivation
§ Scenarios with intermittent connectivity even in urban environments (e.g. high cost,
fading, closed APs), and:
§ Personal wireless devices with significant storage capability
§ majority of applications are related to data gathering
§ Data exchange in challenge networking scenarios
§ performance improvement by exploiting social interactions and structure
§ users are not interested in knowing the location of data
IntroductionMotivation & Goal
SCORP: Social-aware Content-based Opportunistic Routing
Content knowledge (i.e., content type, interested parties)
✚
Social proximity
ê
Faster, better content reachability in challenged networks
3
Problem Statement How to exploit social proximity and content knowledge to augment the efficiency of opportunistic wireless networks?
IntroductionProblem Statement & Known Solutions
Opportunistic forwarding: content-oblivious solutions
Opportunistic forwarding: content-aware solutions
Approach Properties SCORP: Similarities of Differences
CiPRO Classifies type of contact based on users’ interactions SCORP does not aim to predict future encounters.
Bubble Rap Combines node centrality with community structures Similar to SCORP : • dLife weighs the levels of social interaction
between nodes and computes their importance • Bubble Rap: uses social interactions to identify
communities and popular nodes.
dLife Uses users' behavior found in their daily life routines
Approach Properties However, SCORP is independent from:
SocialCast Captures the node's future co-location with others sharing the same interest
Connectivity degree and node co-location
ContentPlace Considering the user's social strength towards the different communities that he/she belongs to and/or has interacted with.
Content availability, and users' communities.
Content-oblivious vs Content-Awareness: not clear the advantages of content-awareness in terms of the data dissemination in challenge networks
4
Probability of encountering nodes with a certain content interest among the ones with similar daily social habits
SCORP: Social-aware Content-based Opportunistic Routing Utility Function & Algorithm
CD(a,b1)CD(a,c2)
CD(a,d3)CD(a,e4)
CD(a,f1)
08:00a.m. 04:00p.m.12:00p.m. 08:00p.m. 12:00a.m. 04:00a.m. 08:00a.m.Daily Sample Ti
AA1
2A
1
2A
1
2
W(a,1)i 1
CA
3
24
1A
3
24
1A
3
24
W(a,2)i
CD(a,b1)CD(a,b1)
CD(a,b1) CD(a,b1)CD(a,c2) CD(a,c2)
Time-Evolving Contact to Interest x :
5
SCORP Evaluation
Scenario
§ Opportunistic Network Environment (ONE) § Synthetic mobility model:
§ 12-day interaction in the city of Helsinki § 150 nodes divided into 8 groups of people and 9
groups of vehicles. § Human mobility traces (Crawdad):
§ 36 nodes in Cambridge University § 2 months § 32 sporadic contacts per hour
§ Traffic § Message size: from 1 to 100 KB. § Buffer space: 2 MB § Time-To-Live (TTL): 1, 2, 4 days, 1, and 3 weeks
Benchmarks
§ dLife:
§ 24 daily samples of one hour
§ Bubble Rap:
§ K-clique for community detection
§ Spray and Wait (serves as lower bound in what concerns delivery cost):
§ L = 10
Results analysed in terms of:
§ Average delivery probability: ratio between delivered messages and messages that should have been delivered
§ Average cost: number of replicas per delivered message
§ Average latency: time elapsed between message creation and delivery
Major Findings
§ Bubble Rap: affected by the absence of central nodes (only 20% in this set). Creates the highest number of replicas
§ dLife: 21% better than Bubble Rap, but leads to useless replications due to high number of contacts. Generates 64.5% to 65.2% less replicas than Bubble Rap
§ Spray and Wait 58.6% and 37.7% better than Bubble Rap and dLife: nodes cover most of the simulated area
§ SCORP: 64.7%, 44.5%, and 10.7% better than Bubble Rap, dLife and Spray and Wait. Creates up to 99.4% less replicas than dLife
§ Latency: SCORP has a subtle advantage over Spray and Wait and dLife (up to 6.4% and 17.6% less latency, respectively) for short lived messages.
6
SCORP Evaluation of TTL Impact on a synthetic model
0
200
400
600
800
1000
1200
1400
1600
1 day 2 day 4 day 1 week 3 week
#of
repl
icas
TTL
Average CostSpray andWait
Bubble RapdLife
SCORP
0
0.2
0.4
0.6
0.8
1
1 day 2 day 4 day 1 week 3 week
%
TTL
Average Delivery Probability
Spray andWaitBubble Rap
dLifeSCORP
5!1031!104
2!104
3!104
4!104
1 day 2 day 4 day 1 week 3 week
Sec
onds
TTL
Average Latency
Spray andWaitBubble Rap
dLifeSCORP
Major Findings § Spray and Wait has low performance (nodes follow routines and do not cover the whole simulated area) and low cost/high latency (few nodes are used to forward).
§ Bubble Rap has low performance due to buffer exhaustion; highest cost due to community creation.
§ dLife and SCORP have similar behaviour: dLife may lead to buffer exhaustion (approximately 24% more than the allowed)
§ SCORP: keeps resource usage at a low usage rate. Experiences up to 93.61%, 90.25% and 89.94% less latency than Spray and Wait, Bubble Rap and dLife
7
SCORP Evaluation of Network Load Impact based on traces
0
0.2
0.4
0.6
0.8
1
1 5 10 20 35
%
# of messages/interests per node
Average Delivery Probability (1-day TTL)
Spray andWaitBubble Rap
dLifeSCORP 5!103
1!104
2!104
3!104
4!104
1 5 10 20 35
Sec
onds
# of messages/interests per node
Average Latency (1-day TTL)
Spray andWaitBubble Rap
dLifeSCORP
0
20
40
60
80
100
1 5 10 20 35
#of
repl
icas
# of messages/interests per node
Average Cost (1-day TTL)Spray andWait
Bubble RapdLife
SCORP
Conclusions § Efficiency of data dissemination over challenged networks can be improved when forwarding is designed having content knowledge and social proximity in mind
§ SCORP has better performance than previous social-aware content-oblivious routing proposals:
§ Delivers up to 97% of its content in an average of 46.9 minutes
§ Bubble Rap needs 335.5 minutes and dLife and 343.7 minutes
§ SCORP produces up to approximately 13.9 and 4.7 times less replicas than Bubble Rap and dLife
8
SCORP Conclusions and Future Work
Future Work § Implement SCORP as content dissemination application in challenge networking scenarios: Amazon region in Brazil
§ Show the advantages of SCORP in relation to other content-oriented social-aware solutions (SocialCast , ContentPlace and CiPRO ) as soon as the code of such approaches is made available,