Peer Matcher: Decentralized Partnership Formation

Abstract

Abstract-This paper presents PeerMatcher, a fully decentralized algorithm solving the k-clique matching problem. The aim of k-clique matching is to cluster a set of nodes having pairwise weights into k-size groups of maximal total weight. Since solving the problem requires exponential time, PeerMatcher employs a novel set of heuristics that aim at converging to the optimal grouping while keeping the associated time and computational complexity low. A key feature is the use of peerto-peer communication. An extensive evaluation of PeerMatcher demonstrates its accuracy, efficiency, and scalability.

RideMatcher: Peer-to-Peer Matching of Passengers for Efficient Ridesharing

“…Our system is based on our earlier PeerMatcher system [10]. However, as stated in section II, there are several changes that have to be made in order to adapt PeerMatcher to the ridesharing problem:…”

“…This is particularly difficult in a distributed setting, where matching operations have to be synchronized in order to avoid inconsistencies. We address this challenge by extending PeerMatcher, our previous protocol for distributed partnership formation [10]. With PeerMatcher it is possible to partition a weighted graph into groups of a fixed size, such that the weights of the groups are maximal.…”

“…We take as starting point our previous research on a peerto-peer algorithm for solving the k-clique matching problem [10]. The goal of this problem is to find all non-intersecting k-cliques in a weighted graph, such that all the cliques have maximal weight.…”

“…1. Swapping in PeerMatcher [10] in the graph is represented by a ride, each weighted edge represents the similarity between two rides, and k is the number of passengers each car can carry. Therefore, we study the possibility of adapting our earlier PeerMatcher algorithm [10] to the ride matching problem.…”

“…Swapping in PeerMatcher [10] in the graph is represented by a ride, each weighted edge represents the similarity between two rides, and k is the number of passengers each car can carry. Therefore, we study the possibility of adapting our earlier PeerMatcher algorithm [10] to the ride matching problem. We identify three main changes that have to be made: i) we need to find a way for nodes to connect with other nodes upon entering the system, since in PeerMatcher this is done artificially at initialization ii) we have to define a weight function that assesses the similarity between car rides, and iii) since cars can have different capacities, we have to make the algorithm work with cliques of different sizes.…”