The power of memory in randomized broadcasting

• Authors: Robert Elsässer;Thomas Sauerwald
• Venue: Proceedings of the nineteenth annual ACM-SIAM symposium on Discrete algorithms
• Year: 2008

In this paper we analyze the runtime and number of message transmissions generated by simple randomized broadcasting algorithms in random-like networks, and show that an apparently minor change in the ability of the nodes implies an exponential decrease in the average communication overhead produced by these algorithms at an arbitrary node. A natural randomized broadcasting protocol, called random phone call model, has been introduced by Karp et al. [20]. In this model it is assumed that in each time step, every node of G calls a neighbor, chosen uniformly at random, and establishes a communication channel with this neighbor. Any node u is then allowed to send/receive messages to/from all nodes which have established communication channels with u in the current time step. Karp et al. showed that some piece of information r, placed initially on one of the nodes of a complete graph of size n, can be spread with probability 1 - n-Ω(1) to all nodes of this graph within O(log n) time steps, by using O(n log log n) transmissions related to r. Furthermore, they proved that this result is asymptotically optimal among all address-oblivious broadcasting algorithms. In a recent paper [9], we analyzed the random phone call model in traditional random graphs Gn,p with p log2 n/n, and showed that any address-oblivious broadcasting algorithm requires with high probability Ω(log n) time steps and Ω(n (log log n + log n/log(pn))) message transmissions to inform all nodes of such a graph. In this paper we consider two simple modifications of the random phone call model, and show that in both cases the number of total message transmissions can be reduced significantly (up to almost a logarithmic factor). In the first case, we allow each node of a random graph Gn,p with p logδ n/n, where λ is a properly chosen constant, to call in every time step four different neighbors, chosen uniformly at random, and we prove that the number of message transmissions decreases to O(n log log n). This can be viewed as a "power of multiple choices" type theorem for randomized broadcasting. Then we show that if in the random phone call model the nodes are provided with a little memory, i.e., they are able to remember the addresses of the nodes chosen in the most recent three time steps, then the communication overhead decreases substantially, too. Finally, we prove the optimality of our results. The algorithms presented in this paper can cope with restricted communication failures, only require rough estimates of the number of nodes, and are robust against slight topological changes. In addition, our results can be extended to the generalized random graph model of [6].