Parallelization of Entity-Based Models in Computational Social Science: A Hardware Perspective

Abstract

The use of simulations by social scientists in exploring theories and hypotheses is well documented. As computer systems have grown in capacity, so have interests of social scientists in executing larger simulations. Social scientists often approach their simulation design from the top down by selecting an Entity-Based Model (EBM) framework from those that are readily available, thus limiting modeling capability to the available frameworks. Ultimately, the framework is dependent upon what is at the bottom, the hardware architecture that serves as the foundation of the computing system. Parallel hardware architecture supports the simultaneous execution of a problem split into multiple pieces. Thus, the problem is solved faster in parallel. In this thesis, a selection of parallel hardware architectures is examined with a goal of providing support for EBMs. The hardware's capability to support parallelization of EBMs is described and contrasted. A simple EBM is tested to illustrate these capabilities and implementation challenges specific to parallel hardware are explored. The results of this research offer social scientists better informed choices than the sequential EBM frameworks that currently exist. Matching the model to the correct supporting hardware will permit larger scale problems to be examined and expands the range of models that a social scientist can explore.