Using an Improved Data Structure in Hybrid Memory for Agent-Based Simulation

Abstract

Data structure is an important issue to get good performance in parallel and distributed applications. These data structures have to be designed with the memory paradigm in mind where the data structure will be used in order to explore the architecture in a better way and subsequently obtain the best Speedup. Current parallel programming languages enable us to easily transform a parallel solution developed for a distributed paradigm to a hybrid solution just by adding pragma codes. At first approach, this is an interesting solution because it does not require several code modifications. Nevertheless, this interchange can cause a slowdown if an appropriate and deep adaptation is not carried out in the code. In this paper, we present our experience when we migrated a data structure developed for a distributed paradigm to a hybrid paradigm. This data structure was implemented in our Fish Schooling Agent-Based simulator where it might be useful either as a distributed paradigm or a hybrid paradigm. The results show the importance of customizing the data structure for the appropriate infrastructure and parallel programming paradigm. We believe that the data structure should have a flexible and dynamic behavior in accordance with the paradigm used.