Web service discovery is currently performed with centralized registries such as UDDI. In this paper, we propose a super-peer network protocol to combine the efficiency of a centralized protocols and P2P networks. For avoiding a flooding the network with search request and for minimizing the number of messages routed in the network, we represented content-addressable network (CAN) structure, which provides a scalable, fault-tolerant distributed hash table (DHT), for super-peers communication. Web service definitions implemented semantically as OWL ontology. The proposed architecture offers self-maintaining and self-clustering network where the peer groups classify the Web service definitions and each peer-group becomes the owner of a classification dynamically.