Automotive and avionics embedded systems are usually composed of several tasks that are subject to complex timing constraints. In this context, the LET paradigm was introduced to improve the determinism of a system of tasks that communicate data through shared variables. The age latency corresponds to the maximum time for the propagation of data in these systems. Its precise evaluation is an important and challenging question for the design of these systems. We consider in this paper a set of multi-periodic tasks that communicate data following the LET paradigm. Our main contribution is the development of mathematical and algorithmic tools to model precisely the dependency between tasks executions to experiment with an original methodology for computing the age latency of the system. These tools allow to handle the whole graph instead of particular chains and to extract automatically the critical parts of the graph. Experiments on randomly generated graphs indicate that systems with up to 90 periodic tasks and a hyperperiod bounded by 100 can be handled within a reasonable amount of time.