Presentation is loading. Please wait.

Presentation is loading. Please wait.

Dept. of Computer Science Distributed Computing Group Asymptotically Optimal Mobile Ad-Hoc Routing Fabian Kuhn Roger Wattenhofer Aaron Zollinger.

Similar presentations


Presentation on theme: "Dept. of Computer Science Distributed Computing Group Asymptotically Optimal Mobile Ad-Hoc Routing Fabian Kuhn Roger Wattenhofer Aaron Zollinger."— Presentation transcript:

1 Dept. of Computer Science Distributed Computing Group Asymptotically Optimal Mobile Ad-Hoc Routing Fabian Kuhn Roger Wattenhofer Aaron Zollinger

2 Distributed Computing Group Overview Introduction Model Face Routing Adaptive Face Routing Lower Bound Conclusion

3 Distributed Computing Group Model I (Ad-Hoc Network) Nodes are points in R 2 All nodes have the same transmission range (normalized to 1) ) network is a unit disk graph Distance between any two nodes is lowerbounded by a constant d 0 )  (1) -model

4 Distributed Computing Group Model II (Geometric Routing) Nodes know the geometric positions of themselves and of their neighbors Source s knows the coordinates of destination t Nodes not allowed to store anything In the message, only O(logn) additional bits can be stored

5 Distributed Computing Group Cost Model I Cost of sending a message over a link (edge) e is c(e) Cost of a path p is the sum over the costs of its edges Cost of a routing algorithm A is the sum over the costs of the traversed edges

6 Distributed Computing Group Cost Model II 3 different cost metrics: Link distance metric ( c (e) ´ 1 ) Euclidean distance ( c d (e) ) Energy metric ( c E (e):=c d 2 (e) ) more general: c E (e):=c d  (e) for an  ¸ 2

7 Distributed Computing Group Costs are Equivalent Lemma: In the  (1) -model the link, Euclidean, and energy metrics of a path or an algorithm are equivalent up to a constant factor.

8 Distributed Computing Group Face Routing geometric routing algorithm for planar graphs Compass Routing on Geometric Networks [Kranakis, Singh, Urrutia; 1999]

9 Distributed Computing Group Face Routing

10 Distributed Computing Group Face Routing (Faces)

11 Distributed Computing Group Face Routing st

12 Distributed Computing Group Face Routing st

13 Distributed Computing Group Face Routing st

14 Distributed Computing Group Face Routing st

15 Distributed Computing Group Face Routing st

16 Distributed Computing Group Face Routing st

17 Distributed Computing Group Face Routing st

18 Distributed Computing Group Face Routing st

19 Distributed Computing Group Face Routing st

20 Distributed Computing Group Face Routing st

21 Distributed Computing Group Face Routing st

22 Distributed Computing Group Face Routing st

23 Distributed Computing Group Face Routing st

24 Distributed Computing Group Face Routing st

25 Distributed Computing Group Face Routing st

26 Distributed Computing Group Face Routing st

27 Distributed Computing Group Face Routing st

28 Distributed Computing Group Face Routing st

29 Distributed Computing Group Face Routing st

30 Distributed Computing Group Face Routing st

31 Distributed Computing Group Face Routing st

32 Distributed Computing Group Face Routing st

33 Distributed Computing Group Face Routing st

34 Distributed Computing Group Face Routing st

35 Distributed Computing Group Face Routing st

36 Distributed Computing Group Face Routing st

37 Distributed Computing Group Face Routing (Analysis) Lemma: Face Routing always finds a path to the destination. The total cost of Face Routing is O(n).

38 Distributed Computing Group Face Routing (Analysis) Proof Sketch: each face is explored at most once each edge is traversed at most four times ) there are at most 3n-6 edges (Euler formula)

39 Distributed Computing Group Problem of Face Routing Face Routing always reaches destination However, even if source and destination are close to each other, Face Routing can take O(n) steps. We would like to have an algorithm, whose cost is a function of the cost of an optimal path.

40 Distributed Computing Group Adaptive Face Routing st

41 Distributed Computing Group Adaptive Face Routing st

42 Distributed Computing Group Adaptive Face Routing st

43 Distributed Computing Group Adaptive Face Routing st

44 Distributed Computing Group Adaptive Face Routing st

45 Distributed Computing Group Adaptive Face Routing st

46 Distributed Computing Group Adaptive Face Routing st

47 Distributed Computing Group Adaptive Face Routing st

48 Distributed Computing Group Adaptive Face Routing st

49 Distributed Computing Group Adaptive Face Routing st

50 Distributed Computing Group Adaptive Face Routing st Ellipse is too small, go back!

51 Distributed Computing Group Adaptive Face Routing st

52 Distributed Computing Group Adaptive Face Routing st

53 Distributed Computing Group Adaptive Face Routing st

54 Distributed Computing Group Adaptive Face Routing st

55 Distributed Computing Group Adaptive Face Routing st

56 Distributed Computing Group Adaptive Face Routing st

57 Distributed Computing Group AFR on the Unit Disk Graph AFR needs a planar graph, UDG is not planar need a planar subgraph of UDG: –simple distributed construction –spanner for link distance, Euclidean, and energy metric

58 Distributed Computing Group Two nodes u and v are connected by an edge iff the circle with uv as diameter contains no other node. Gabriel Graph Definition: u v

59 Distributed Computing Group Properties of GG Å UDG For each pair of nodes, the GG ( Å UDG) contains an energy optimal path (on UDG). ) spanner for link, Eucl. dist., energy (  (1) -model) planar no additional communication ) meets all our requirements

60 Distributed Computing Group AFR Complexity Theorem 1: Let c * be the cost (link, Euclidean, or energy) of an optimal path between two nodes on the UDG. Applying AFR on GG Å UDG then terminates with cost O(c * 2 ).

61 Distributed Computing Group AFR Complexity, Proof I Lemma: For each used ellipse E, the cost is linear in the number of nodes in E. Collorary: In the  (1) -model, for each used ellipse E, the cost is linear in area covered by E.

62 Distributed Computing Group AFR Complexity, Proof II Lemma: The cost, AFR needs to route a packet, is linear in the area covered by the last used ellipse. Ellipses grow exponentially 

63 Distributed Computing Group AFR Complexity, Proof III Lemma: Using an ellipse E, AFR finds a path from s to t iff there is such a path inside E. Lemma: All paths of (Euclidean) length smaller or equal to c are inside an ellipse whose area is in O(c 2 ).

64 Distributed Computing Group AFR Complexity, Proof IV All the lemmas together now prove Theorem 1.

65 Distributed Computing Group Lower Bound Theorem 2: Let c * be the cost (link, Euclidean, or energy) of an optimal path between two nodes on a UDG G. For each geometric routing algorithm, there is a graph for which the cost is  (c * 2 ).

66 Distributed Computing Group Lower Bound, Proof w t s Cost of optimal path: Exp. cost for an algorithm A :

67 Distributed Computing Group Main Theorem Theorem 3: On the Unit Disk Graph in the  (1) -model, AFR is asymptotically optimal. (follows directly from Theorems 1 and 2)

68 Distributed Computing Group Conclusion I works fine for link and Euclidean distance (still  (c * 2 ) ) For energy, it can be shown that the cost of a geometric routing alg. cannot be bounded by a function of c * alone.  (1) restriction can be dropped by clustering:

69 Distributed Computing Group Conclusion II The lower bound holds for all routing algorithms which have only local knowledge at the beginning. if coordinates of dest. are not known, but if each node can store some bits: Then there is a simple flooding variant which achieves O(c * 2 ).


Download ppt "Dept. of Computer Science Distributed Computing Group Asymptotically Optimal Mobile Ad-Hoc Routing Fabian Kuhn Roger Wattenhofer Aaron Zollinger."

Similar presentations


Ads by Google