Network topology information of peer-to-peer applications is useful for examining their characteristics as well as defining an accurate model to simulate their behaviors. Unfortunately, dynamic behaviors of peers such as join and leave make it difficult to obtain the accurate network topology information. In this paper, we propose a method to accurately measure the topology of a Winny network even under heavy churns by observing search queries through a geographically distributed crawling system. After obtaining topology information from the crawling system, we attempt to identify important nodes, the removal of which would partition the network as an example characterization of the Winny network. We evaluate which metric such as node degree, betweenness centrality, and conductivity efficiently identifies such important nodes.