Parallel simulation models generally take one of two approaches: optimistic or conservative. The optimistic approach allows the model to reach an unacceptable state, then rolls back to an acceptable state after realizing that it has reached an unacceptable state. The rollback is commonly achieved by using a technique called state saving.

The authors propose reverse computation, which performs the inverses of the individual operations that led to the unacceptable state, as a new rollback technique. Using two examples, ATM Multiplexor Cascade and PCS Network, the authors demonstrate that their new approach has insignificant forward computation overhead and low state memory requirements in fine-grain simulations; achieves better caching, resulting in two- to three-fold speedup when compared to copy state-saving, periodic state-saving, and incremental state-saving; and can be automated using compiler-based techniques.

The paper is well written and self-contained. Every researcher or practitioner working on parallel simulation should read it.