Batch based cancellation: a rollback optimal cancellation scheme in time warp simulations

An efficient cancellation scheme is essential to the performance of Time Warp simulations. The pitfalls of rollback echoes, chasing hazards and cascading rollbacks can be identified as being attributable to the inefficiency of the conventional per-event based cancellation scheme. Instead of capturing the happen-before relation between events, which is used by the range based cancellation scheme, the batch based cancellation scheme proposed in this paper utilizes a modified paradigm of vector time, namely, state vector, to capture the dependence of events. We prove that with conformance to specific rules regulating the advancement of LPs, the events to be cancelled by a straggler message can be determined using a range of the state vector. Thus, knowledge of the range enables any LP to recover from the receipt of a straggler message at the cost of at most one rollback (i.e., rollback optimal). The results of preliminary experiments conducted using a manufacturing model show that the proposed scheme is successful in reducing the number of anti-messages and increasing the ratio of the number of committed events to the number of processed events.