Using consistent global checkpoints to synchronize processes in distributed simulation

Abstract

This paper presents a discussion about using consistent global checkpoints to synchronize processes of a program of distributed simulation during the rollback procedure, allowing to improve the simulation performance and to carry out a more suitable memory management. A new optimistic protocol is presented as consequence of using consistent global checkpoints.

Distributed simulation: state-of-the-art and potential for operational research

“…Many of these focus on reducing the number of rollbacks and cascades, the need to reprocess event messages (for example when an arriving event message does not change the simulation state) and more efficient methods of calculating GVT. Examples include lazy cancellation (Gafni, 1988), lazy re-evaluation (Fujimoto, 1990), restricted rollback (Damani, Wang, & Garg, 1997), adaptive state saving (Rönngren & Ayani, 1994;Lin, et al 1993), message aggregation (Chetlur, et al 1998), global checkpointing (Moreira, Santana, & Santana, 2005) and speculative computing (Venu & Joe, 2014). Some attempts have also been made to control unconstrained rollback by combining optimistic and conservative approaches (e.g.…”

Distributed Building Evacuation Simulator for Smart Emergency Management

“…Another approach is to use message aggregation techniques in an effort to optimize the communication process among the agents [42]. In [43], the authors discuss the option of using consistent global checkpoints to synchronize the processes of a distributed simulation during the rollback procedure, allowing to improve the simulation performance and to carry out a more suitable memory management. They present a new optimistic protocol that uses consistent global checkpoints.…”

A Lightweight Heuristic-based Mechanism for Collecting Committed Consistent Global States in Optimistic Simulation

“…Hence approaches based on independent checkpointing activities of the LPs, as we have assumed in our proposal, remain the most effective ones in general applicative scenarios. Also, compared to the work in [15], our solution addresses the issue of exploiting consistency of global states not for synchronization purposes, but for the evaluation of global predicates in synergy with GVT advancement.…”

“…To the best of our knowledge, there has been a single proposal based on consistent global checkpoints in the context of optimistic simulation systems [15], which uses control information piggybacked on application messages in order to track the state dependencies and direct forced checkpoints to construct consistent global snapshots to be exploited for synchronization purposes. However, this has been shown to provide adequate performance only for reduced simulation model sizes.…”