Agent-based simulation (ABS) is a paradigm which has received much attention within the last years. For enabling industrial use of ABS scalable solutions are required which can be executed on a distributed computing architecture. Such solutions must be capable of simulating complex models with hundreds or more complex deliberative agents with really autonomous behavior. In this article we argue that only optimistic synchronization protocols are potentially capable of providing the required performance. We suggest a synchronization protocol with constrained optimism which exploits specific characteristics of communication patterns within agent based simulations. Furthermore the presented protocol includes appropriate methods for GVT computation and fossil collection in distributed ABS as well as mechanisms to ensure repeatability.