1. Scalable Routing In Delay Tolerant Networks
Mohammad Reza Faghani

2. An Outer Space Network station Jupiter Mars Mars Mars Earth
Can not have direct access

3. What is Delay Tolerant Networks ?
Intermittent link (dis)connection
No guarantees on End-to-End path
Frequent long duration partitioning C B A D E F

4. What is Delay Tolerant Networks ?
High latency, low data rate
Order of hours latency
Long queuing times
Because of disconnections (Store and Forward)
Extremely large (hours, even days)
Constraints on end node
Limited power, limited buffer
How packets route !?

5. Routing in DTN DTN Routing Challenges.
Instantaneous end to end path may not exist.
Large queuing delays.
Buffer limitations at intermediate nodes.
Large messages.

6. Routing in DTN Routing Goals: Eventual Delivery (delivery ratio)
Minimizing delivery delay
Scalability
Cost-effective

7. The Routing Problem in DTN
Nodes with finite storage capacity
Links with dynamic behavior
Time varying capacity (c(t))
C(t) = 0, if link is down
Message
Src, Dst, start-time, size
Output: Compute path(s) for every message
Objective: Minimize delay
Other objectives: message delivery ratio, minimize $$ cost

8. The Routing Problem in DTN
Edge parameterized by
Source
Destination
capacity function
delay function.
Define link costs and find minimum cost path.
Cost varies with time.
Compute minimum cost paths over this dynamic cost assignment
Modified Dijkstra by taking into account time of arrival

9. Input variable used in Routing
Contacts
Complete link time variant datas
Contacts summary :
Time independent information
Average waiting time until the next contact
Queuing : Link queues, available storage
Traffic Demand

10. Routing input vs. Performance
Input variable used
“Performance”
Contacts
Summary +
Local
Queuing
MED ED
EDLQ
EDAQ LP
distributed
Global
Traffic
Demand

11. Routing in large networks
As the network size grows, number of contacts increases.
These algorithms are not scalable for large networks.
Using the idea used in static scalable routing.
The hierarchical routing

12. Scalable Routing in DTNs
Cong et. al. proposed a simple DTN model.
This makes hierarchical routing possible
For scalability, defined two contact information compression methods.

13. Simplified DTN model Static nodes (white)
Mobile nodes with repetitive motion
Motion cycle:
T1=2 mins, T3=T4=3 mins
Contact: a time period for communication.
Persistent contacts: (2,1), (3,4), & (5,6)
Persistent contact: (ni nj - - -)
Predicted contacts: (1,3), (3,6), & (4,5)
Predicted contact: (ni nj Tij tstart tduration) 6 5 1 3 2 4
As a sample sattelite networks, define contacts by their category, define 0.1 , 0.01

14. Hierarchical Routing in Static Networks5 1 4 18 19 20 21 11 10 12 22 23
Hierarchical network
Uses multilevel clustering.
Offers scalable management of routing tables.
Hierarchical routing
Uses the hierarchical network as a topology abstraction
A top-down process: the decision made in a higher level is more important
Clustering & Clusterhead 8 2 6 13 24 14 15

15. Hierarchical Clustering in Static Networks
Level 2 2 3 1 2 7 3 6 4 5
Level 1
Level 0 10 25 11 12 1 1 1 15 19 16 24 2 2 2 7 7 7 22 17 9 13 18 20 21 8 23 3 3 3 6 6 6 14 4 4 4 5 5 5

16. Hierarchical Clustering in Static Networks1
Level 2 2 3 1 2 7 3 6 4 5
Level 1
Level 0 10 11 25 12 1 1 15 19 16 24 2 2 7 7 22 17 9 13 18 20 21 8 23 3 3 6 6 14 4 4 5 5

17. Hierarchical Clustering in Static Networks
Before any routing each node in the network needs to obtain the topology information of its clusters in all levels.
Source should know the hierarchy address of destination.

18. Hierarchical Clustering in Static Networks1
Level 2 2 3 1 2 7 3 6 4 5
Node 61 represents the cluster of nodes 60,80,240
All nodes have their own hierarchy address
e.g. node 6 HA equals (13, 22, 61, 80).
Level 1
Level 0 10 11 25 12 1 1 15 19 16 24 2 2 7 7 22 17 9 13 18 20 21 8 23 3 3 6 6 14 4 4 5 5 18

19. Hierarchical Routing in Static Networks1
Level 2 2 3 1 2 7 3 6 4 5
Level 1
Level 0 10 11 16 12 1 1 15 19 16 24 2 2 7 7 22 17 9 13 18 20 21 8 23 3 3 6 6 14 4 4 5 5
Destination
Source 19

20. Hierarchical Routing in DTNs
Similar to that in static networks
Multilevel clustering
Clusterhead selection
Links: contact information aggregation
Contact information compression methods

21. Cluster head Selection
Objective
Clusterhead: the center (in terms of delay) of a cluster
Cluster members are close to their clusterheads
Absolute priority
D(i,j) is the weighed average delay between nodes i and j
Higher if n is closer to the shortest paths among its neighbors
Clusterheads that have the highest APs are self-selected.
Relative priority
Node i selects a nearby clusterhead n who has a high AP

22. Contact information aggregation
Hierarchical links have time-variant delays
They contain aggregate contact information
Contact information in a level k+1 link are aggregated from the related level k links. 6 5 1 3 2 4

23. Contact information compression
Aggregation level (La )
Above La, each link contain only a constant delay

24. Contact information Compression
Contact information aggregated to link (6,7) is shown in (c).
There are two possible shortest paths across the time as shown in (d) & (e)
The contact information stored by link (6,7) after contact information removal

25. Hierarchical Routing in DTNs
Similar to Hierarchical Routing in static networks
Hop by hop routing
A top-down decision making within each hop
Step 1: top-down routing
When the routing process is above La
Step 2: Routing with contact information
Routes on the combined contact information in all clusters below La

26. Simulation An example network Mobile node Static nodes Mobile nodes
Whose trajectory travels several random waypoints within a random square bound

27. Simulation Results
Route length Distribution

28. Simulation Results
Hop-count ratio
70/30

29. Simulation Results
Delay ratio
70/30

30. Simulation Results
DHR+CIR Delay ratio
70/50

31. Simulation Results
Storage communication overhead

32. Summary
Summary
Routing performance is close to the optimal routing result in terms of hop-count and delay
Routing performance improves as aggregation level (La ) increases
Routing performance improves as the source and destination distance increases
Storage and communication overhead is reduced by the compression methods while desirable routing performance and scalability is achieved.

33. References
[1] Liu C,Wu Jie. Scalable Routing in Delay Tolerant Networks.In proc. of the 8th ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc), 2007
[2] S. Jain, K.Fall, and R.Patra. Routing in Delay Tolerant networks. In Proc. of ACM SIGCOMM, 2004
[3] Leonard Kleinrock, Farok Kamoun, "Hierarchical Routing for Large Networks, Performance Evaluation and Optimization", Computer Networks, Vol. 1, No. 3, pp. 155–174, January 1977

34. Questions !?