How to improve the performance of content searching in peer-to-peer (P2P) systems is a challenging issue. In this paper we attack this problem by proposing a new decentralized P2P architecture ? ontology-based community overlays. The system exploits the semantic property of the content in the network to cluster nodes sharing similar interest together to improve the query and searching performance. Specifically, a distributed hash table (DHT) based overlay is constructed to assist peers organizing into communities. Those peers in the same community form a Gnutella-like unstructured overlay. This architecture helps reduce the search time and decrease the network traffic by minimizing the number of messages propagated in the system. Moreover, it retains the desirable properties of existing unstructured architectures, including being fully decentralized with loose structure, and supporting complex queries. We demonstrate by simulation, that with this architecture, peers can get more relevant resources faster and with less traffic generated.