Diversity of Forwarding Paths in Pocket Switched Networks

1
Diversity of Forwarding Paths in Pocket Switched
Networks

• Vijay Erramilli
• with
• Mark Crovella,
• Augustin Chaintreau,
• Christophe Diot

2
Pocket Switched NetworksWhat are they?

• Challenge Providing connectivity in the absence
  of infrastructure
• PSNs seek to exploit huge amount of untapped
  resources in portable devices
• Local wireless b/w, Storage, CPUs
• In a PSN, messages are propagated using all
  available opportunities network connection or
  user mobility

3
PSNs Illustrated
Message ( Source, Destination, Time of
origin) Paths Series of successive contacts from
source to destination
j
D
S
Message originates at time t
i
4
Communicating in such settings

• How to communicate in such settings?
• Dynamic underlying topology-stable paths not
  present
• Standard routing inappropriate
• Instead, need to devise an effective strategy
  for forwarding
• Forwarding strategy depends strongly on setting

• Specific setting we are interested in
• Finite of nodes
• Bounded physical space -
• High node mobility

5
To study forwarding, study paths

• We hope to learn how to do effective forwarding
  by studying the properties of available paths

• Our Contributions
• Investigate available paths and characterize
  them
• Use characteristics of available paths to
  study forwarding

6
Measurement Driven Approach

• Contact data from conference settings
• Infocom, Conext 2006, 100 Nodes, 4 days, high
  contact density
• Trace driven simulations Choose source and
  destination node randomly
• 3 hour periods from traces
• Chosen for stability

Infocom 06 15-18
7
(No Transcript)
8
(No Transcript)
9
Explaining Path Explosion

• Analytical Model
• Homogenous contact rates ?, Poisson contacts
• Model evolution of Sn(t) Number of paths at
  node n at time t
• Number of paths (in expectation)
• Exponential explosion in of paths agrees with
  empirical results

..but does not explain all facets of observed
Path Explosion relation between T1, TE
10
Heterogeneity of Contact Rates
Conext 06
Infocom 06
To investigate effect of heterogeneity, we label
nodes Nodes with rates gt Median rate In
High Mobility Nodes with rates lt Median rate
Out Low Mobility
11
Heterogeneity of Contact Rates Helps
Explain Observed Phenomena

In-In
In-Out
Out-In
Out-Out
11
12
So far we have seen

• Path explosion phenomenon Soon after optimal
  path arrives at destination, many paths arrive
• Shown empirically
• Developed analytical model to help explain
• Better explained using heterogeneous rates of
  contact
• How does this help us understand forwarding?

13
Impact on Forwarding

• Trace driven simulations of a wide array of
  forwarding algorithms
• Metrics
• Success rate
• Avg. delay (averaged over all delivered messages)
• Cost - not yet

14
Spectrum of Algorithms

• Epidemic Forward Always - Baseline
• Destination Aware
• Fresh Forward to node if it sees destination
  more recently
• Greedy Forward to node if it sees destination
  more times
• Greedy Online Online version of above
• Dynamic Programming Generalization of Fresh
• Destination Agnostic
• Greedy Total Forward to node if it has more
  total contacts

15
(No Transcript)
16
Paths taken by different algorithms
Path Explosion helps explain similarity in
performance
16
17
Possible Implications

• Higher contact rate nodes faster path explosion
• Suggest strategy forward to high contact rate
  nodes

18

• Thanks!

19
(No Transcript)
20
Pocket Switched Networks

• Make use of local/global/any connectivity to
  propagate messages Be opportunistic!
• Under more generic topics
• MANETs
• DTNs

20
21
Problem Setting

• Abstract Problem
• Finite Mobile Nodes, Restricted Physical Space,
  High Mobility
• Source Node Ns, Destination Node Nd
• Path Sequence of tuples (Ns, t1), (Ni, t2),
  ..,(Nd,tk)
• Goals of Forwarding Algorithm
• Find a path to Node Nd if exists reaching at
  time tk
• Find it in shortest possible time (tk-t1)