Deferred vs. immediate modification of simulation state in a parallel discrete event simulator using threaded worker pools

The joint integrated mission model (JIMM) is a real-time legacy battlefield simulator employed in detailed analyses and virtual exercises. To leverage more processors to improve real-time execution, a worker pool of threads optimistically executes events in parallel but avoids cascading rollback by executing only one future event per simulated object. Safeguards for maintenance of simulation state are programmed explicitly and either deferred or immediate modification of state variables could be employed in case of event rollback. In the beginning of the main parallelization effort, deferred modification was used where simulation state is updated only when the event can be completed safely. However, after successful implementation, it was determined to be impractical. Later, all safeguard programming employed immediate modification where original state is restored in case of rollback. This paper discusses these techniques for parallel execution of events in JIMM from initial efforts through later code maintenance.