Predictive Adaptable Logger (PAL)

Large federated simulations are inherently complex to design and execute effectively to meet experiment objectives. When running a series of computationally expensive simulations, simulation engineers need to quickly analyze critical events of interest (e.g., time of detection) without being inundated with GBs of irrelevant data. Unfortunately, many critical events are sporadic, so to ensure that no relevant data are missed, data are currently logged at a high fidelity and frequency throughout the simulation run, resulting in up to 250 GB of data for a federated simulation. This amount of data unnecessarily consumes network resources because only a small percentage of the data is relevant to the critical events.We propose a Predictive Adaptable Logger (PAL) that adapts logging settings based on the current likelihood of a pending critical event and the current cost of failing to log data relevant to a critical event of interest. PAL is initialized on historical simulation runs, so it learns what predictive events forecast a critical event. It also reasons about the cost of a missed critical event based on the progress of the simulation run. PAL will minimize the consumption of resources during federated simulation runs while capturing key data needed for analysis.Approved for Public Release | 17-MDA-9395(24 Oct 17)