The evacuation of whole cities or even regions is an important problem, as demonstrated by recent events such as the evacuation of Houston in the case of Hurricane Rita, or the evacuation of coastal cities in the case of tsunamis. A robust and flexible simulation framework for such large scale disasters helps to predict the evacuation process. Furthermore, it is possible to recognize bottlenecks in advance, so that an elimination of those bottlenecks is possible. This should lead to a better preparedness for cities or regions that face a high risk of natural disasters. Existing methods are either geared towards smaller problems (e.g. Cellular Automata techniques or methods based on differential equations) or are not microscopic (e.g. methods based on dynamic traffic assignment). Our work uses a technique that is both microscopic and capable to process large problems. The simulation is applied to the Indonesian city of Padang. The city faces a high risk of being inundated by an earth quake triggered tsunami.