Abstract. The modularity of Service-oriented Architectures (SOA) allows to establish complex distributed systems comprising e.g., services, clients, brokers, and workflow engines. A growing complexity, however, automatically increases the number of potential fault sources which have effects on the whole SOA. Fault handling mechanisms must be applied in order to achieve a certain level of robustness. In this paper we do not deal with fault-tolerance itself but regard the problem from a different perspective: how can fault-tolerance be evaluated? We argue that this can be best done by testing the system at runtime and observing its reaction on occuring faults. Though, engineers are facing the problem of how to perform such tests in a realistic manner in order to get meaningful results. As our contribution to this issue we present an approach for generating fault injection testbeds for SOA. Our framework allows to model testbeds and program their behavior, to generate running instances out of it, and to inject diverse types of faults. The strength of our approach lies in the customizability of the testbeds and the ability to program the fault-injecting mechanisms in a convenient manner. 1

lb ~heion or the rev n of itmcantentein any mznnerto an uneutho “ lrum b prohibid b> law. InformdiotlW3c ed m be&mM&eddonJ.. ‘. to ~nc in the milje’ry and rm United*tee, a