1. Kevin Lee & Adam Piechowicz 10/10/2009
Delay Tolerant Mobility Aware Routing/Mobility Dissemination Protocol for the Airborne Network
Kevin Lee & Adam Piechowicz
10/10/2009

2. Problem Statement An end-to-end path is not always guaranteed
Packets have to be delivered in a delay-tolerant fashion
How to use planned AN backbones’ trajectories to deliver packets to minimize:
Packet failure rate
Delay
Local buffer

3. Contributions
Provide best next forwarding hop in the delay tolerant network based on current network condition
Provide congestion avoidance and load balancing by local queuing awareness mechanism

4. MARP/MDP+DTN Design Think of the topology as a time-varying graph
We can select the best next hop given a specified metric (minimum delay)
Use modified Dijkstra’s algorithm with time-varying edge costs
w(e(u,v), t) indicates the cost of using edge (u, v) after time time t
The cost is predominantly the time after t that (u, v) is up
Propagation delay negligible
Transmission rate not considered
Will say something about the disadvantage of this scheme (unawareness of congestion), but will mention it’s not so bad at all because it compares favorably with all-knowledge oracles

5. Example 1 Source node at 1 and each edge cost is 1 N Node 1, Pre
{1,2,3,4}
0,1 ∞
{2,3,4}
1,1
5,1
{3,4}
5 < 1 + w(e(2,3), 1) = 1 + (4+1) = 6
1 < 1 + w(e(2,4),1) = 1 + (2 + 1) = 4
{3} {}
Node 1, Pre: Node 1’s delay and its predecessor.

6. Example 2 N Node 1, Pre Node 2, Pre Node 3, Pre Node 4, Pre
{1,2,3,4,5}
0,1
1,1 ∞
7,1
4,1
{2,3,4,5}
2,2
{3,4,5}
7 > 2 + w(e(3,4),2) = 2 + (0 + 1) = 3,3
{5}
3,3
4 < 3 + w(e(4,5), 3) = 3 + ∞
{0}

7. Predecessor Computation
Take Node 3 from Example 2
2, 2 means:
It will take Node 1 to reach Node 3 with cost of 2 (the first 2)
The predecessor of Node 3 is Node 2 (the second 2)
One can then trace back to get the complete route traversal and the time at which the packet should be sent

8. Local Queuing Aware Scheduling
Network disconnectivity increases queuing delay
Queuing delay increases congestion
Route around congestion by considering neighbors’ queue size
w(e, L[u] + T) will incorporate:
The cost of sending packets already in the queue plus,
The cost of sending the last packet scheduled to use u to deliver to v

9. Importance of Encountering for Queuing Awareness
The more frequently a node encounters with another node, the more packets the node can offload to that node
Intuitively, the link between these two nodes provide lower delay to the destination
Two types of encountering:
Single
Multiple

10. Single Encountering
A node A single encounters another node B if node A meets node B only once in a period
Where time(1,e,L[u]+T) indicates the time Node u meets Node v at which time one packet in Node u’s queue is delivered since L[u] + T
P is the period of the time-varying graph
Qsize is the queue size at Node u

11. Single Queuing Example
Assume there are 2 messages in Node 2’s queue N
Node 1, Pre
Node 2, Pre
Node 3, Pre
Node 4, Pre
{1,2,3,4}
0,1 ∞
{2,3,4}
1,1
5,1
{3,4}
5 < 1 + w(e(2,3), 1) = = 18
1 < 1 + w(e(2,4),1) = = 16
{3} {}
Node 1, Pre: Node 1’s delay and its predecessor.

12. Single Queuing Example (cont.)
w(e(2,3), 1) = because it takes
5 more seconds to dequeue the first packet,
6 seconds to dequeue the second packet,
another 6 seconds to dequeue the last packet
By the eqn, w(e(2,3), 1) = * 2 = 17
Packets in Node 2’s queue will use the same edge in consideration, e.g.,
w(e(2,3),1) considers first two packets going to Node 3
w(e(2,4),1) considers first two packets go to Node 4

13. Multiple Encountering
A node A multiple encounters another node B if node A meets node B more than once in a period
Tx(e,P-(L[u]+T)): the number of times e is up during P-(L[u]+T time
Tx(e,t) is the # of times e is up during the remaining time t of one period

14. Multiple-Encounter Queuing Example
Assume there are 2 messages in Node 2’s queue N
Node 1, Pre
Node 2, Pre
Node 3, Pre
Node 4, Pre
{1,2,3,4}
0,1 ∞
{2,3,4}
1,1
5,1
{3,4}
5 < 1 + w(e(2,3), 1) = = 10
1 < 1 + w(e(2,4),1) = = 16
{3} {}
Node 1, Pre: Node 1’s delay and its predecessor.

15. Multiple-Encounter Queuing Example (cont.)
w(e(2,3), 1) = because it takes
3 more seconds to dequeue the first packet,
2 more seconds to dequeue the second packet,
4 seconds to dequeue the last packet
Since Qsize (=3) > Tx (=2), eqn (2) is used:
w(e(2,3), 1) = (6 – 1) = 10

16. Handling Multiple-Traffic Flows
Contacts (the time-varying at any given point in time) is known
Local queuing is known; approximate global queuing by keeping track of messages along each routing path
Traffic demand is known,
It is a set of messages
Each message is a tuple (u,v,t,m), where u is the source of the msg, v is the destination, t is the time the msg is sent, m is size
Buffer constraints are given
THE ORACLE HAS COMPLETE KNOWLEDGE! – A linear programming exercise

17. Approximate Optimality
LP is computationally expensive! – Computation become too large for practical example
Use contacts and queuing oracle (EDAQ) instead
“EDAQ compares favorably, in terms of average delay, with the optimal solution.”
However,
Global knowledge may not be required for good performance in many cases
Implementing the queuing oracle, in particular, may not be worthwhile
EDAQ: Earlier delivery with all queues

18. Approximate Optimality (contd.)
Contact oracles (ED) might just be enough for our scenarios!
Lesson: TOO MUCH KNOWLEDGE MAY NOT ALWAYS BE GOOD!
Didn’t God say not to eat the wrong tree?

19. Evaluation

20. MARP/MDP vs. MARP/MDP+DTN
Network Flow from GlobalHawk to AWACs 2
Solid arrow shows the desired network flow
Dotted lines shows current available connections
3 experimental variables
Radio range
Delay tolerance
Flow volume

21. Results
Delay tolerance represents largest improvement in packet delivery, 52% (col1 &2)
Fixed range: Low flow has higher delivery and lower latency (col 1 & 3)
Fixed flow: High radio range has higher delivery ratio and lower latency than low radio range

22. MARP/MDP+DTN vs. MARP/MDP+DTN+QC
Two separate rate flows from 1 and 2 to 5
Transmission rates range from Mbps to 1024 kbps

23. Results
When transmission rate is high, PDR for MARP/MDP+DTN is 0%

24. Packet Difference between Node 3 and 4
Both flow forward to Node 3 heavily in MARP/MDP+DTN
MARP/MDP+DTN+QC is able to divert traffic and achieve load balancing

25. Latency Delay extremely long in MARP/MDP+DTN
Result indicates the need for local queuing awareness

27. Conclusion & Future Work
MARP/MDP+DTN shows the benefit of delay tolerance
MARP/MDP+DTN+QC shows the benefit of local queuing awareness
Congestion scenario configuration to verify local queuing aware scheduling
Tune parameters/routing metrics of MARP/MDP+DTN+QC protocol in accordance with flight and link data obtained from real flight tests like Capstone II