An Eecient Conditional-knowledge Based Optimistic Simulation Scheme

We present in this paper a single-hop version of the conditional-knowledge approach to optimistic simulations 10]. The algorithm has potentially lower memory requirements than the conventional time-warp algorithm and uses a single negative message between processes for cancellation of multiple erroneously sent messages. In time-warp, each processed message is stored both on the output queue of a process and on the input queue of another process. In our algorithm, output queues are not needed because causal dependency between events is maintained by the receivers. A process upon rollback, simply sends a single negative message to its immediate output neighbors who then use the message to cancel appropriate events from their input queues. To permit correct cancellation at the receiver, each regular message carries only a single tuple describing the assumptions made in the generation of the message. Performance results indicate that the algorithm compares favorably with conventional time-warp.