The aim of P2P computing is to build virtual computing systems dedicated to large-scale computational problems. This kind of architecture will contain several thousands of computing nodes which need to be organized. Moreover, the architecture will deal with node's dynamism and network scalability. JXTA proposes an underlying infrastructure which build a virtual network of peers on top of a physical network. This structure allows, among other features, to create peer groups. JNGI, which is one of the first P2P decentralized computing platforms, is already using these groups to structure the computing system. This article aims to demonstrate how to build new types of groups called similarity groups into the JNGI project. A similarity group is a peer group where all the peers have common caracteristics like CPU speed or memory size. This is done in order to increase the relevance of task dispatching and therefore to increase the performance of JNGI.