RTI中乐观推进机制的实现

Implementation of Optimistic Advancing Mechanism in RTI

正确理解乐观推进机制的基本原理,并按照高层体系结构HLA(highlevelarchitecture)规范实现RTI(runtimeinfrastructure)中的乐观推进服务一直是分布式仿真领域关注的难点问题.介绍了并行离散事件仿真PDES(paralleldiscreteeventsimulation)和HLA中的乐观推进机制,并指出了它们之间的重要差异,例如PDES中的虚拟时间(virtualtime)可以回卷(rollback),回卷发生在进程中;而HLA中的逻辑时间不能够回卷,但乐观盟员在不影响保守盟员推进的情况下可以回卷自己调度事件的时间,回卷发生在盟员内而不是RTI内.另外,提出了在实现乐观推进服务时,不需要RTI作任何保存操作的零保存技术,并成功地将该技术应用到RTI软件StarLink中.零保存技术通过在消息句柄类型RTI::MessageRetractionHandle定义中增加两个变量,一个表示TSO(timestamporder)消息的时标,另一个表示所有接收该消息的盟员(RTI::MessageRetractionHandle为IEEE1516.1定义的数据类型),RTI将具有此类型的消息句柄返回给发送TSO消息的盟员保存,当发送盟员再次使用消息句柄撤消(retract)消息时,RTI从消息句柄中就可以知道并通知接收盟员撤消消息.对于理解和开发RTI中的乐观推进服务具有重要的现实意义.

In distributed modeling and simulation area, it is intractable to comprehend the fundamental principles of optimistic advancing mechanism and implement the optimistic advancing services in RTI (runtime infrastructure) according to HLA (high level architecture) specifications. This paper introduces two different optimistic advancing mechanisms in PDES (parallel discrete event simulation) and HLA, and reveals some important differences between them. For example, the virtual time can be rolled back in PDES, the logical time can not be rolled back in HLA, and an optimistic federate can only roll back its message-scheduling time and must ensure that its rollback won't influence the advancing of conservative federates. Rollback occurs in a logical process in PDES, but it can only occur in a federate rather than RTI in HLA. An implementation mechanism called Zero-Saving is proposed. With this mechanism, a RTI does not need to save any execution states when optimistic advancing services are implemented. This mechanism has successfully been applied to a RTI named StarLink. The Zero-Saving mechanism adds two new variables into the message retraction handle type RTI: Message Retraction Handle which is a data type defined by IEEE 1516.1. One variable represents the time stamp of the sent TSO (time stamp order) message, and the other is used to save all federates which receive the message. When a TSO message is sent to RTI, RTI returns the sending federate a message retraction handle with all message-received federates. So RTI knows which federates should be notified to retract the received messages whenever the sending federate uses a message retraction handle to ask RTI to retract a TSO message. The fundamental principles and implementation of optimistic advancing services introduced are useful for RTI developers.