Optimistic parallel discrete event simulation uses a state history to support the rollback operation when a causality error occurs. While the frequent state saving increases the state saving overheads, the infrequent state saving strategy increases the coast forward cost. Therefore, an analytical framework is needed to keep the recovery cost low. In this talk we propose a cost model to minimize the recovery cost by using a probabilistic approach to determine which state to be saved. Implementation results indicate that the cost model reduces the elapsed times by 35% and reduces the coast forward and state saving cost by 44% as compared to saving-every-state approach.