Node clustering is an efficient way to improve query hit radio and decreases network traffic cost in unstructured P2P topology. In this paper, we study the relationship between the similarity of peers properties and system clustering coefficient. It demonstrates that increase peers-similarity is an efficient way to increase system-clustering coefficient. By this theorem, we have proposed three algorithms base on unstructured P2P topology to improve the peers-similarity. The experiments show the algorithms will help P2P system reflect the small-world characteristics and reduce the average time complexity of per lookup request.