Synchronization techniques in parallel discrete-event simulation are categorized as conservative and optimistic. Conservative synchronization techniques can be categorized into “synchronous” and “asynchronous” approaches. In this research paper, the authors describe a composite synchronization method that combines the synchronous and asynchronous conservative techniques. In the composite synchronization technique, the model is partitioned into synchronous and asynchronous parts. Either type of corresponding technique will be applied when the other leads to an inefficient simulation. The method has been implemented in the Dartmouth scalable simulation framework, with the intention to make it possible for non-specialists in the field to use parallel discrete-event simulation in the communication networks domain.
The partition of the model into synchronous and asynchronous is in fact an optimization problem. In this paper, a theoretical optimization model is described. This model is a sound theoretical basis for composite synchronization. The implementation, however, searches at runtime for the conditionally optimal partition whose measured behavior appears to optimize performance.
The paper concludes with the results of experiments using a synthetic simulation model, and a model of a network. Both experiments show that the composite technique performed better than when only the synchronous or asynchronous methods were applied.
This is a clear and well-written paper. Specialists in the field will not encounter problems in understanding the issues raised in this paper. Non-specialists, however, will need more background information. The paper provides references to the necessary theory.