In unstructured P2P file sharing systems, the ununiform distribution of file popularity causes damage to the usability of the system. The search for rare files has more chance to fail. Flooding algorithm with indices which are exchanged randomly based on gossip protocol provides a success guarantee in a probability sense. However, the problem of low success rate and large numbers of messages still exist for queries of rare files. In this paper, we propose an index scheme and IBFS, a search algorithm based on nodepsilas information capacity. In contrast to the former, our method reduces 50%bandwidth consumption on average and performs efficient object discovery, making the system more stable and available.