The virtual region-based P2P architecture was provided in this paper, which extended the proximity concept in overlay network. By virtual region it meant various related factors should be integrated into the design of P2P topology, such as peer interest, peer locality, and heterogeneity in peer capacity etc. But, the shared contents in P2P are vast and miscellaneous, it is very difficult to recognize the interest similarity among peers, dynamically form the interest region and direct peer to appropriate interest region. In this paper, the Jensen-Shannon Divergence (JSD) was used to characterize interest/semantic similarity among peers, and the interest region formation and location mechanism based on Dynamic Interest Landmark (DIL) was offered to facilitate to dynamically guide peers to join the appropriate interest region corresponding to peers' content semantic. Considering that interest region is composed of geographically sporadic peers, each interest region was organized according to peer proximity in underlying infrastructure. The theoretical and simulated results illustrated that our architecture can significantly reduce the query overhead, and improve the search performance.