Traditional recommender systems rely on the use of a central server. This server is a single point of failure; moreover, the need for an expensive central infrastructure may lead to economic dependencies, forming an incentive to manipulate recommendations. Decentralized recommender systems on the basis of peer-to-peer networks can help solving this problem. Most existing approaches, however, are not scalable for large networks, or they do not consider security and privacy issues. The article at hand presents a peer-to-peer recommender system based on the chord overlay network and item-based collaborative filtering. Scalability is improved by introducing various optimizations. Additionally, we present an approach that reduces the potential for manipulation by system participants while granting a high degree of privacy.