Efficient Routing in Networks with Long Range Contacts

  • Authors:
  • Lali Barrière;Pierre Fraigniaud;Evangelos Kranakis;Danny Krizanc

  • Affiliations:
  • -;-;-;-

  • Venue:
  • DISC '01 Proceedings of the 15th International Conference on Distributed Computing
  • Year:
  • 2001

Quantified Score

Hi-index 0.00

Visualization

Abstract

We investigate the notion of Long Range Contact graphs. Roughly speaking, such a graph is defined by (1) an underlying network topology G, and (2) one (or possibly more) extra link connecting every node u to a "long distance" neighbor, called the long range contact of u. This extra link represents the a priori knowledge that a node has about far nodes and is set up randomly according to some probability distributions p. To illustrate the claim that Long Range Contact graphs are a good model for the small world phenomenon, we study greedy routing in these graphs. Greedy routing is the distributed routing protocol in which a node u makes use of its long range contact to progress toward a target, if this contact is closer to the target, than the other neighbors. We give upper and lower bounds on greedy routing on the n-node ring Cn augmented with links chosen using the r-harmonic distributions. In particular, we show a tight 驴(log2 n)-bound for the expected number of steps required for routing in Cn augmented using the 1-harmonic distribution. Hence, our study shows that the model of Kleinberg [11] can be simplified by using the ring rather than the mesh while preserving the main features of the model. Our study also demonstrates the significant difference (in term of both diameter and routing) between the ring augmented with long range contacts chosen with the harmonic distribution and the ring augmented with a random matching as introduced by Bollobas and Chung [3]. Finally, using epimorphisms of a graph onto another, for any network G, we show how to define a probability distribution p and study the performance of greedy routing in G augmented with p. For appropriate embeddings (if they exist), this performance turns out to be O(log2 n).