Task-parallel in-situ temporal compression of large-scale computational fluid dynamics data

Abstract

Present day computational fluid dynamics simulations generate extremely large amounts of data, sometimes on the order of TB/s. Often, a significant fraction of this data is discarded because current storage systems are unable to keep pace. To address this, data compression algorithms can be applied to data arrays containing flow quantities of interest to reduce the overall amount of storage. Compression methods either exactly reconstruct the original dataset (lossless compression) or provide an approximate representation of the original dataset (lossy compression). The matrix column interpolative decomposition (ID) can be implemented as a type of lossy compression for data matrices that factors the original data matrix into a product of two smaller factor matrices. One of these matrices consists of a subset of the columns of the original data matrix, while the other is a coefficient matrix which approximates the columns of the original data matrix as linear combinations of the selected columns. Motivating this work is the observation that the structure of ID algorithms makes them a natural fit for the asynchronous nature of task-based parallelism; they are able to operate independently on sub-domains of the system of interest and, as a result, provide varied levels of compression. Using the task-based Legion programming model, a single-pass ID algorithm (SPID) for CFD applications is implemented. Performance studies, scalability, and the accuracy of the compression algorithms are presented for an analytical Taylor-Green vortex problem, followed by a large-scale implementation of a compressible Taylor-Green vortex using a high-order Navier-Stokes solver. In both cases, compression factors exceeding 100 are achieved with relative errors at or below 10e-3. Moreover, strong and weak scaling results demonstrate that introducing SPID to solvers leads to negligible increases in runtime.