Towards correct distributed simulation of high-level Petri nets with fine-grained partitioning

Abstract

Powerful grid and cluster computers allow efficient distributed simulation. Optimistic simulation techniques have been developed which allow for more parallelism in the local simulations than conservative methods. However, they may require costly rollbacks in simulation time due to dependencies between model parts that cause violations of global causality. Different notions of time have been proposed to detect and remedy these situations. Logical time (or Lamport time) is used in many present-day distributed simulation algorithms. However, high-level colored Petri nets may contain global activity priorities, vanishing states, and global state dependencies. Thus virtual time is not sufficient to maintain the global chronological order of events for the optimistic simulation of this model class. The paper presents a new approach that guarantees a correct ordering of global states in a distributed Petri net simulation. A priority-enhanced vector time algorithm is used to detect causal dependencies. © Springer-Verlag Berlin Heidelberg 2004.