A general framework and integrated methodology towards scalable heterogenous computations for structural dynamics on massively parallel platforms

A general framework and a unified integrated computational technology encompassing a wide variety of new and existing time integration operators within the scope of Linear Multi-Step (LMS) methods is now possible employing a single analysis code via a unified family of generalized integration operators [GInO] towards scalable computations for structural dynamics on massively parallel computing platforms. A unified scalable computational approach towards such a computational technology is desirable for large-scale structures and large processor counts. This present paper proposes the recent developments of a unified scalable approach for non-linear and linear structural dynamics which inherits three critical scalability properties: numerical scalability, parallel scalability and computer memory utilization scalability whilst simultaneously providing a variety of choices to the analyst. The numerical scalability analysis is conducted via an integrated unified technology for large deformation, elastic, elastic-plastic dynamic response. For geometric non-linearity a total Lagrangian formulation and for material non-linearity elasto-plastic formulations are employed. The other key distinguishing features are the development of coarse-grained parallel computational models via generalized time integration operators that be can ported to a wide range of parallel architectures using a message-passing paradigm (using MPI), graph partitioning techniques, and domain decomposition techniques. This is the first time that such a capability is plausible via a unified technology and the developments further enhance computational structural dynamics areas.