On Exploiting Transient Contact Patterns for Data Forwarding in Delay Tolerant Networks
Wei Gao and Guohong Cao
Dept. of Computer Science and EngineeringPennsylvania State University
Outline
IntroductionTrace-based Pattern FormulationData Forwarding ApproachPerformance EvaluationSummary & Future Work
Data Forwarding in DTNs Carry-and-Forward methods
Mobile nodes physically carry data as relaysMajor problem: appropriate data forwarding metric for
relay selectionData forwarding metric measures the node’s capability of
contacting others in the future
B
A C
0.7
0.5
Our FocusEffective data forwarding metric for a short time
constraintTransient node contact pattern
Consider different time periods in a dayNode contact pattern may vary temporallyRelays selected based on cumulative contact patterns
may not be the best choice during a short time period
Transient Contact Pattern Transient contact distribution
Highly skewed during different time periods Example: different people contact during different time periods
in a day Cannot be differentiated from the cumulative contact rate
Contact frequency
Transient Contact Pattern Transient connectivity
Some nodes remain connected during specific time periods to form Transient Connected Subnets (TCS)
Example: a student remains connected with his classmates during the class
Multi-hop communication within the TCS
TCS
10:00AM
11:00AM
Transient Contact Pattern Node A sends data to a destination at 12pm with a 6-hour
time constraint Valid time period: 12pm-18pm A node forwards data to another node with higher data
forwarding metric
Cumulative contact rate Transient
contact rate
Direct contact
Transient connected subnet
Major ContributionsFormulate transient node contact patterns based
on realistic DTN tracesDevelop data forwarding metrics to analytically
predict node contact capability with better accuracy
Outline
IntroductionTrace-based Pattern FormulationData Forwarding ApproachPerformance EvaluationSummary & Future Work
TracesRecord user contacts at university campusVarious wireless interfaces
Bluetooth: periodically detect nearby peersWiFi: associate to the best Access Point (AP)
Transient Contact DistributionSkewed distribution of node contacts
Over 50% between 12pm and 16pm
Less than 7% between 22pm and 7am
Transient Contact Distribution Alternative appearances of on-period and off-period
Contact process during on-periods is stable and predictableContacts during off-periods are random and unpredictable
On-periodAn on-period is determined by a set of contacts
happened at time . For ,
EndStart
Distribution of On/Off-Period LengthAccurately approximated by normal distribution
hours for both traces
On-periods are short
Transient ConnectivityDistribution of contact duration
Many contacts with non-negligible duration
Over 20% last longer than 1 hour
Transient Connectivity The temporal change of TCS size of a mobile node
Formulated as Gaussian function
Outline
IntroductionTrace-based Pattern FormulationData Forwarding ApproachPerformance EvaluationSummary & Future Work
Data Forwarding Metric The capability Ci of node i to contact other nodes
during time tc: the current time te: the data expiration timeThe expected number of nodes that node i can contact
N: the total number of nodes in the networkcij: pairwise contact probability between node i and j
Pairwise Contact Probability Basic idea: only based on the contact process during
on-periods
Case 1: tc is within anon-going on-period
Case 2: the next on-period starts before te
Case 1: tc is within an on-going on-period
, where
Pairwise Contact Probability
Contact process within an on-period is modeled as Poisson
PDF of the on-period length
T = te – tc
Case 2: the next on-period starts before te
, where
Pairwise Contact Probability
The last on-period ends before tc
PDF of the off-period length
Exploiting Transient ConnectivityA node indirectly contacts all the nodes in a TCS
if it contacts any one node in the TCS
Temporal change of TCS size:
Outline
IntroductionTrace-based Pattern FormulationData Forwarding ApproachPerformance EvaluationSummary & Future Work
ComparisonsData forwarding metrics
Contact Counts (CC)BetweennessCumulative Contact Probability (CCP)
Data forwarding strategiesCompare-and-ForwardDelegation ForwardingSpray-and-Wait
Performance ComparisonCompare-and-Forward in the UCSD trace
100% 20%Better delivery ratio
Similar forwarding cost
Two Cases of Contacts
70%
30%
SummaryEffective data forwarding metrics for data
forwarding in DTNs with a short time constraintImproves delivery ratio with similar forwarding cost
Transient node contact patternsTransient contact distributionTransient connectivity
Future workThe overhead of identifying transient contact patternThe temporal evolution of social network structure
Thank you!
http://mcn.cse.psu.edu
The paper and slides are also available at:http://www.cse.psu.edu/~wxg139
Length Distribution of On/Off-Period
On-periods are generally shortMost contacts are covered by on-periodsWe have hours for both traces
Characterizing Transient Contact Pattern
Transient contact patterns are characterized at real-time at each nodeFor nodes i and j, the parameters of their on/off-period
are updated every time they contactEach node detects its TCS whenever it contacts
another node A detecting beacon message is broadcasted within the TCS Each node in the TCS replies to the original sender upon
receiving the beacon.
Prediction Error
The start of a new on-period?Random contact in an off-period?
Node contact randomnessUnclassifiable contact
Off-periods longer than 24 hours Most contacts happen during on-periods
The two cases are only found at nodes with low contact frequency
Different values of Ton
8 8
Node Buffer ConstraintForwarding only one data item
A node simply drops the data in cases of limited bufferThe consideration of node buffer constraint is trivial
Forwarding multiple data itemsThe consideration of node buffer constraint is
formulated as a knapsack problemVarious data forwarding metrics can be applied
Related Work Node contact capability is estimated from various ways
Prediction of node mobility and co-location events Semi-Markov chain, Kalman filter
Node contact pattern as abstraction of node mobilityExploitation of social network concepts
Centrality, social community, homophily Various metrics apply to the same forwarding strategy
Single-copyMultiple-copy
Spray-and-Wait, Delegation Forwarding