Well known system models of distributed systems assume that all processes are already running when the algorithm starts. This includes the assumption that there are enough non-faulty processes in the system to run an algorithm correctly in the presence of failures. This diploma thesis deals with the problem of initializing a network where the processes are down at the beginning and therefore don't receive messages until they have booted. When a process is up it has to decide when it is save to start the desired service, based only on the information it gets from messages. That includes that the process has no timing information. The analysis of the algorithm is done using a partially synchronous system model where exist lower and upper bounds on the message transmission delay which are not known by the processes. We consider the problem of round synchronization. The solution is a modification of the well known non authenticated broadcast primitive by Srikanth and Toueg.