Multi-attribute resource discovery in structured Peer-to-Peer (P2P) networks is still an open problem.Recently

a novel and more efficient multi-attribute resource discovery approach than traditional ways — P2P-based Intelligent Resource Discovery (PIRD) has been proposed.However

PIRD may be inefficient under network churns.To address this issue

in this paper

a gossip-based PIRD (G-PIRD) is proposed.G-PIRD employs a gossip algorithm to learn the estimated value of network size and dynamically publishes resource indexes to keep a great efficiency of resource discovery.Meanwhile

a load balancing scheme based on bounded LSH (B-LSH) is proposed to deal with the problem of potential load imbalancing of G-PIRD.Extensive experiments show that G-PIRD can adapt well to the changes of network size to maintain high query efficiency and the proposed load balancing scheme can greatly reduce the maximum number of indexes per node with the slight loss of query efficiency.