A Performance Prediction Model for Tomographic Reconstruction in Structural Biology

Abstract

Three-dimensional (3D) studies of complex biological specimens at subcellular levels have been possible thanks to electron tomography, image processing and 3D reconstruction techniques. In order to meet computing requirements demanded by the reconstruction of large volumes, parallelization strategies with domain decomposition have been applied. Although this combination has already proved to be well suited for electron tomography of biological specimens, a performance prediction model still has not been derived. Such a model would allow further knowledge of the parallel application, and predict its behavior under different parameters or hardware platforms. This paper describes an analytical performance prediction model for BPTomo – a parallel distributed application for tomographic reconstruction-. The application's behavior is analyzed step by step to create an analytical formulation of the problem. The model is validated by comparison of the predicted times for representative datasets with computation times measured in a PC's cluster. The model is shown to be quite accurate with a deviation between experimental and predicted times lower than 10%.