Trigger Happy: Assessing the Viability of Trigger-Based In Situ Analysis

Triggers are an emerging strategy for optimizing execution time for in situ analysis. However, their performance characteristics are complex, making it difficult to decide if a particular trigger-based approach is viable. With this study, we propose a cost model for trigger-based in situ analysis that can assess viability, and we also validate the model's efficacy. Then, once the cost model is established, we apply the model to inform the space of viable approaches, considering variation in simulation code, trigger techniques, and analyses, as well as trigger inspection and fire rates. Real-world values are needed both to validate the model and to use the model to inform the space of viable approaches. We obtain these values by surveying science application teams and by performing runs as large as 2,040 GPUs and 32 billion cells.