To enhance the efficiency of resource discovery and distribution, it is necessary to effectively estimate the network performance, in terms of metrics such as round-trip time and hop-count, between any peer-pair in a peer-to-peer application. Measurement in large scale network for peer-to-peer applications and services faces many challenges, such as scalability and quality of service (QoS) support. To address these issues, we present an architecture of a group-based performance measurement service (gMeasure), which disseminates and estimates network performance information for peer-to-peer applications. The key concept in gMeasure is Measurement Groups (MeGroups), in which hosts are self-organized to form a scalable hierarchical structure. A set of heuristic algorithms are proposed to optimize the group organization to reduce system cost and improve system measurement accuracy.