Assessment of a Black-Box Approach for a Parallel Finite Elements Solver in Computational Hemodynamics

Abstract

Numerical approximation of blood flow has emerged in the last 20 years as a tool to investigate physiopathology of the circulation, moving from a proof-of-concept to a clinical stage. By merging medical images with numerical models it is possible to support the decision-making process of surgeons and doctors in general. In particular, the iCardioCloud project aims at establishing a framework to perform a complete patient-specific hemodynamics analysis for aortic diseases such as dissections, occlusions and aneurysms. From a computer science standpoint, such a project faces multiple challenges. First of all the dimension of the problem in terms of number of equations to be solved for each patient is in general huge and thus it requires massively parallel methods. In addition, clinical timeline demands for efficiency, since availability of results -- at least in an emergency scenario -- should be granted in few hours from data retrieval. Therefore it is mandatory to develop a good implementation on high-end parallel systems, such as large clusters or even supercomputers. Unfortunately, it is not straightforward to obtain an efficient implementation on such machines. In this paper we discuss a parallel implementation obtained with a black-box approach, that is set up by assembling existing packages and libraries and in particular LifeV, a finite element library developed for Computational Fluid Dynamics. The ultimate goal is to assess if the application can be solved efficiently and which is the parallel paradigm that best matches the computational requirements.