Parallel Implementation Details of OPENATOM

An accurate understanding of phenomena occurring at the quantum scale can be achieved by considering a model representing the electronic structure of the atoms involved. The Car-Parrinello ab initio Molecular Dynamics (CPAIMD) method [1,2,3,4] is one such algorithm which has been widely used to study systems containing $10-10^3$ atoms. The implementation of CPAIMD in CHARM++ is called OPENATOM [5,6]. To achieve a fine-grained parallelization of CPAIMD, computation in OPENATOM is divided into a large number of virtual processors which enables scaling to tens of thousands of processors. We will look at the parallel implementation of this technique, understand its computational phases and the communication involved and then analyze the benefit from topology-aware mapping of its objects.