Trigger detection for adaptive scientific workflows using percentile sampling

TL;DR

This paper introduces a scalable, noise-tolerant trigger detection method using percentile sampling for adaptive scientific workflows, enabling efficient identification of critical simulation phases without high computational costs.

Contribution

The paper presents a novel, scalable percentile-sampling approach with error bounds for trigger detection in complex simulations, reducing computational overhead significantly.

Findings

Effective detection of rapid heat release increases in combustion simulations

Sampling approach maintains accuracy with problem size independence

Negligible overhead compared to full CEMA computations

Abstract

Increasing complexity of scientific simulations and HPC architectures are driving the need for adaptive workflows, where the composition and execution of computational and data manipulation steps dynamically depend on the evolutionary state of the simulation itself. Consider for example, the frequency of data storage. Critical phases of the simulation should be captured with high frequency and with high fidelity for post-analysis, however we cannot afford to retain the same frequency for the full simulation due to the high cost of data movement. We can instead look for triggers, indicators that the simulation will be entering a critical phase and adapt the workflow accordingly. We present a method for detecting triggers and demonstrate its use in direct numerical simulations of turbulent combustion using S3D. We show that chemical explosive mode analysis (CEMA) can be used to devise a noise-tolerant indicator for rapid increase in heat release. However, exhaustive computation of CEMA values dominates the total simulation, thus is prohibitively expensive. To overcome this bottleneck, we propose a quantile-sampling approach. Our algorithm comes with provable error/confidence bounds, as a function of the number of samples. Most importantly, the number of samples is independent of the problem size, thus our proposed algorithm offers perfect scalability. Our experiments on homogeneous charge compression ignition (HCCI) and reactivity controlled compression ignition (RCCI) simulations show that the proposed method can detect rapid increases in heat release, and its computational overhead is negligible. Our results will be used for dynamic workflow decisions about data storage and mesh resolution in future combustion simulations. Proposed framework is generalizable and we detail how it could be applied to a broad class of scientific simulation workflows.