Time Warp State Restoration Via Delta Encoding

Optimistic simulation yields impressive performance gains for many models. State saving is a quick way to provide the rollback mechanism required for this approach, but it has some drawbacks: it may not handle models with massive states or be able to support memory-constrained systems. This work presents a novel approach to state saving by storing only the relative changes caused by an event. Compressing these deltas allows retaining a greater number of noncommitted events and allows Time Warp to further exploit parallelism in a window less constrained by memory limitations. By compressing the data, we realize greater returns in performance and avoid memory limitations on event / state sizes. Compression ratios over 200 are observed; and, despite chosen pathological conditions, state restoration is fast and efficient. Runtimes are often faster using delta encoding than are their conservative counterparts, without the need for complex reverse code or large memory consumption.