Due to resource constraints common in sensor nodes it is often complicated to profile the performance of an application. One solution is simulating the node and profiling the application in there. This talk presents a flexible infrastructure to generate a call graph and calculate the function runtime.