In most existing simulators, the outputs of a simulation run consist either in a simulat ion report generated at the end of the run and summarizing the statistics of interest, or in a (set of) trace file(s) containing raw data samples produced and saved regularly during the run, for later post-processing. In this paper, we address issues related to the management of these data and their on-line processing, such as: (i)~the instrumentation code is mixed in the modeling code; (ii)~the amount of data to be stored may be enormous, and often, a significant part of these data are useless while their collect may consume a significant amount of the computing resources; and (iii)~it is difficult to have confidence in the treatment applied to the data and then make comparisons between studies since each user (model developer) builds its own ad-hoc instrumentation and data processing. In this paper, we propose OSIF, a new component-based instrumentation framework designed to solve the above mentioned issues. OSIF is based on several mature software engineering techniques and frameworks, such as COSMOS, Fractal and its ADL, and AOP.