Ricardo Taborda

One of the key aspects of the ongoing efforts to push the limits on earthquake simulations, from the computational point of view, is I/O performance. At the front end, the lack of ready accessibility of the input data (velocity model) is one of the major factors affecting the efficiency of mesh construction. Unfortunately, the traditional information that the community possesses in this area is in the form of low-performance computational tools. For several years now, we have been using the in-house developed etree-library to construct etree versions of SCEC’s Community Velocity Model (CVM). We call them CVM-Etrees. They are discrete octree-based representations of the original CVM, with properties that greatly enhance the accessibility of the database. Although the construction of CVM-Etrees does not yet directly solve the performance issue by itself, it is a one time effort which results in a high-performance database that does speed up the process of repetitive mesh generations and material model exploration for preand post-processing analysis. This work explains our approach to storing the SCEC-CVM into etrees and examines their advantages and disadvantages compared to the current version of the CVM. We analyze the querying performance for data exploration and mesh construction processes, and the differences induced by the discretization procedure at different points in the Southern California region. We also present the current CVM-Etrees available to the community, which were recently used by different simulation groups as the primary material database source for the PetaSHA’s ShakeOut earthquake scenario.