An assessment of multi-core for a performance prediction model of tomographic reconstruction

Three-dimensional (3D) reconstruction of structures from projection data is essential for helping people in a wide range of areas. Algebraic reconstruction techniques (ART) are iterative procedures for recovering the structure of the 3D objects from projection images. During the seventies, the ART were dismissed due to high-demanding computing requirements. Interesting recent research aims at acquiring experience with parallelization strategies and at demonstrating the effectiveness of the massively parallel processing approach in 3D reconstructions. Multi-core (MC) technology provides new levels of performance and, therefore, it is of paramount importance to make performance predictions. The objectives of this work are both adapting an analytical performance prediction model for the iterative reconstruction techniques (IRT) to a MC environment and finding a process´s CPU affinity that produces the best overall performance. BPTomo⁺ is a parallel distributed application for tomographic reconstruction that uses IRT. Besides, it includes a process´s CPU affinity mask. The analytical performance prediction model is validated by comparison of the estimated times for representative datasets against BPTomo⁺ computation times measured on a MC server. The analytical model is shown to be quite accurate. The percentage of deviation between estimated and measured times is less than 6%.