Peer-to-Peer (P2P) networks are used by millions of people for sharing music files. As these networks become ever more popular, they also serve as an excellent source for Music Information Retrieval (MIR) tasks. This paper reviews the latest MIR studies based on P2P data-sets, and presents a new file sharing data collection system over the Gnutella. We discuss several advantages of P2P based data-sets over some of the more "traditional" data sources, and evaluate the information quality of our data-set in comparison to other data sources (Last.fm, social tags, biography data, and MFCCs). The evaluation is based on an artists similarity task using Partial Order Embedding (POE). We show that a P2P based Collaborative Filtering dataset performs at least as well as "traditional" data-sets, yet maintains some inherent advantages such as scale, availability and additional information features such as ID3 tags and geographical location.