Querying semantically related data sources depends on the ability to map between their schemas. Unfortunately, in most cases matching between schemas is still largely performed manually. As a consequence, semantic integration issues have become a key bottleneck in the deployment of a large scale integration systems (the number of schemas to map is huge). This work deals with automated methods for matching and efficiently generating schema mappings in large scale environments. We propose a level-wise algorithm based on a semantic distance to evaluate similarity between schema nodes. However, in a large scale context, computing the semantic distance for every couple of nodes cannot be done. Thus, we proposed family based heuristics (CHARLIE) in order to efficiently generate mappings. Experiments have shown that our approach is very efficient for large scale integration especially in a super peer based architecture and that it is relevant for real datasets.