Parallelization and vectorization effects on a code simulating a vitreous lattice model with constrained dynamics

Abstract

In this paper we describe the efforts necessary to parallelize and vectorize a code that simulates a vitreous lattice modeling a liquid-glass transition. This model is characterized by a remarkable degree of parallelism: the evolution rules of the lattice depend only on local properties, therefore vectorial and parallel computation could represent a powerful tool when implementing a programfor the system. However, in a real system, as the temperature decreases during the liquid-glass transition, the particles become more and more frozen in their reciprocal positions. This corresponds, when executing the simulation program on the target machine, to a work load decrease along the system evolution. Therefore without a suitable run-time scheduling strategy, the performance of the parallelized and vectorized code gets worse.