基于Markov链的分布式仿真系统可用度研究

Availability Research on Distributed Simulation System Using Markov Chains

根据HLA仿真系统的特点及不同进程对仿真结果的重要程度,提出了分布式仿真系统的故障模型,并将仿真进程分为关键仿真进程和次要仿真进程;根据处理机是否在进行仿真运算,将处理机的计算时间分为有效时间和无效时间,并以此为基础定义了分布式仿真系统的可用度,然后利用Markov链分析了分别采用两级恢复策略和传统容错策略的分布式仿真系统的可用度。结果表明,在发生故障时,在将仿真节点重启时间算入恢复开销的前提下,两级恢复策略会降低系统的可用度,而传统容错策略由于可立即在备份机上进行故障恢复,所以可使系统获得更高的可用度。最后用一组数据验证了该结论的正确性。

According to the characteristics of HLA-based simulation system and influence of different simulation processes on simulation result, the error models for distributed simulation system （DSS） were proposed and the simulation processes were classified into critical part and trivial part. Based on the fact whether processors perform simulation computation, the processor computation time was classified into useful time and useless time, and accordingly the availability of DSS was defined. Besides, considering DSSs that adopt traditional fault-tolerant schema and two-level recovery schema respectively, both of their availability were analyzed with Markov chain. The result shows that when adding the restarting time of failed node to the recovery overhead, the two-level recovery schema will make the system less available whereas the traditional one can make it more available, because in which fault can be recovered immediately on redundancy processors. The correctness of this conclusion is testified through a set of experimental data.