The research (granted by INRIA) described in this dissertation was performed within the ID-Imag laboratory (sponsored by CNRS (UMR 5132) - INRIA - INPG - UJF) whose aim is to use efficiently parallel architectures. Debugging applications for performances and correctness is necessary to exploit efficiently architectures like clusters of workstations. Our approach of performance debugging is based on behavioural observation of program executions, reconstructed from execution traces. The objective of this thesis is to study methods allowing parallel programs execution traces to be recorded easily for a large variety of parallel programming models and interfaces as well as visualization tools. To provide such mechanisms, our tool (called Tumit)...