In this study we show how one can use Fault-Tolerant Units (FTU) in an optimal way to make a TDMA network robust to bursty random perturbations. We consider two possible objectives. If one wants to minimize the probability of losing all replicas of a given message, then the optimal policy is to spread the replicas over time. This is proved using convexity properties of the loss probability. On the contrary if one wants to minimize the probability of losing at least one replica, then the optimal solution is to group all replicas together. This is proved by using majorization techniques. Finally we show how these ideas can be adapted for the TTP/C protocol.