Distributed simulation using hierarchical rollback

• Authors: R. L. Gimarco
• Venue: WSC '89 Proceedings of the 21st conference on Winter simulation
• Year: 1989

As the complexity of computer and communication systems increases, it becomes increasingly difficult to construct and evaluate models of these systems which can be used to study their performance under varying conditions. A modeling technique which is once again gaining in popularity due to its generality and ability to represent systems in varying degrees of detail is discrete event simulation.Constructing a simulation model of a real or proposed system is well understood and many tools are available. The major problem which arises is model evaluation. Due to the complexity and level of detail in a simulation model, it may require an excessive amount of time and computer resources to perform the evaluation. There is an area of current research, though, which is addressing these performance problems: distributed simulation.Distributed simulation has the potential to reduce the often lengthy run-time required for complex discrete event simulation by using multiple, cooperating processors instead of a single processor for model execution. Also, the computer resource requirements can more easily be satisfied by spreading the demand across a set of processors rather than depending on a single processor to furnish the required resources.In this paper a new technique for distributed simulation called hierarchical rollback is presented. Hierarchical rollback employs multiple processors for the evaluation of a single simulation model. Synchronization and coordination of the processors is based on a unique checkpoint/rollback mechanism. It will be shown how hierarchical rollback can be used to execute a hierarchical simulation model on a distributed set of processors.