Empirical Study on Entity Interaction Graph of Large-Scale Parallel Simulations

The entity interaction graph is an important metaphor for understanding the simulation execution of complex systems on parallel computing environment. Current performance tuning techniques often explore interrelated factors affecting performance, but ignore systematic analysis on the structure and behavior of entity interactions. This paper reports an empirical study on the entity interaction graphs of three systems chosen from different domains: Internet models, molecular dynamics, and social dynamics, respectively. The results of complex networks analysis on the entity interaction graphs demonstrate that the heterogeneous distribution of connections and highly clustering are universal in these complex systems. Generally, these properties are not obvious at the system modeling stage. Moreover, mutual information theory is used to measure the "principle of persistence" as the predictability of partitioning on multiple processors. This study facilitates better understanding and quantifying of the interaction complexity and provides implications on performance tuning for parallel simulation of large- scale complex systems.