Bounded recovery in distributed discrete real-time simulations

This thesis proposal defines the problem of recovery in distributed discrete real-time simulations with external actions; real-time simulations with simulation actions in the real world. A problem that these simulations encounter is that they cannot rely on rollback-based recovery (use of checkpoints) for two reasons. First, some actions in the "real world" cannot be undone, and second, the time allowed for recovery tends to be short and bounded. As a result there is a need for some form of error masking for this category of simulations. We propose an infrastructure for these simulations with external actions based on an active distributed real-time database that features replication of the distributed simulation. The degree of replication is based on the dependability requirements of the individual nodes in the simulation. A guideline for how to decompose a distributed real-time simulation into parts with different requirements on the replication protocol is also defined as an interesting topic to investigate further. We introduce the simulation infrastructure "Simulation DeeDS" featuring a replication-based recovery strategy for the category of simulations mentioned. We also show that some information fusion applications are indeed examples of applications that need real-time simulation with external actions and as such can benefit from the proposed infrastructure