Gerenciamento descentralizado de tempo virtual global em simulação distribuída

Abstract

A distributed simulation model can support the cooperation between synchronous and asynchronous components. When components are distributed, additional controls are needed to maintain synchronization between them, and this can be done through the concept of virtual time. Each component has a local virtual time (LVT) defined by the occurrence of events. As a way to keep the synchronization of events between components, a time known by all is necessary, called global virtual time (GVT). The GVT can be defined as the lowest timestamp between messages in transit and the LVTs of the components. In the conservative simulation, the GVT defines a state of progress, preventing events whose time is less than the GVT can be executed. In this way, violations can be avoided. The management of GVT can be done in a centralized or decentralized way. The first approach defines a component as a GVT manager, which receives control information from the participants of the simulation and calculates the global time based on them. This form of management makes control dependent on the central process, which can impose some obstacles. An example of this is the existence of the single point of failure. In the difficulty of communication with the controller process, the GVT calculation can be compromised. The second approach works in a decentralized manner, allowing all components to participate in the global time estimation. The distribution over the GVT control requires additional structures that allow the components to keep records on the simulation participants in order to be able to estimate the global simulation time based on the available local information. This work presents a decentralized strategy for the GVT calculation, based on related solutions, allowing a distribution of the control message load.