Random Walks, Bisections and Gossiping in Circulant Graphs

Abstract

Circulant graphs are regular graphs based on Cayley graphs defined on the Abelian group $\mathbb{Z}_{n}$ . They are popular network topologies that arise in distributed computing. Using number theoretical tools, we first prove two main results for random directed k-regular circulant graphs with n vertices, when n is sufficiently large and k is fixed. First, for any fixed ?>0, n=p prime and L?p 1/k (logp)1+1/k+? , walks of length at most L terminate at every vertex with asymptotically the same probability. Second, for any n, there is a polynomial time algorithm that for almost all undirected 2r-regular circulant graphs finds a vertex bisector and an edge bisector, both of size less than n 1?1/r+o(1). We then prove that the latter result also holds for all (rather than for almost all) 2r-regular circulant graphs with n=p, prime, vertices, while, in general, it does not hold for composite n. Using the bisection results, we provide lower bounds on the number of rounds required by any gossiping algorithms for any n. We introduce generic distributed algorithms to solve the gossip problem in any circulant graphs. We illustrate the efficiency of these algorithms by giving nearly matching upper bounds of the number of rounds required by these algorithms in the vertex-disjoint and the edge-disjoint paths communication models in particular circulant graphs.