Large scale systems are becoming more and more common today. There are many distributed applications emerging that use the capability of world-wide internetworking. Since a lot of applications need insurance of consistency even in the presence of failures, an adequate support for fault-tolerance is necessary. This can be provided by different paradigms and their implementations. Unfortunately, most of these implementations aim only local area networks, whereas our system, called Phoenix, will aim large scale where additional failure types have to be overcome. This paper shows the problems due to large scale, the limits of actual implementations, and our proposition to solve them.