面向社会网络仿真的混合时间管理算法

摘要

并行仿真技术在大规模社会网络研究中扮演着越来越重要的角色,而仿真时间同步机制是决定并行仿真性能的重要因素,如何发掘社会学研究领域的应用特征以设计高效的时间管理算法是当前时间同步机制研究的重要课题.社会网络仿真中个体间行为模式差异较大,难以提取合适的Lookahead以保证保守时间管理策略的高效运行；同时,社会网络仿真中个体交互情况复杂,采用乐观时间管理策略时,容易引起系统内大量级联回滚.本文针对上述问题提出一种基于社区发现的混合时间管理机制,在逻辑进程的粒度上根据网络结构优化选择保守和乐观时间管理策略,从而最大限度地发掘系统的性能.实验结果表明了该算法的有效性.%Parallel Simulation Technique plays a more and more important role in researching large scale social networks. Synchronization is the crucial factor to affect the parallel simulation performance. Nowadays, it is an important topic of synchronization study to exploit application characteristics so as to develop an effective synchronization algorithm. Due to the larger differences of individual behaviors in the large scale social networks, the conservative approach may become limited by sensitivity to looka-head; due to the complex individual interaction, the optimistic approach appears to be prone to cascading rollbacks. In this paper, we propose a hybrid time management approach where a logic process can autonomously choose to be optimistic or conservative, and optimize the selection by utilizing the community detection. The experimental results demonstrate the efficiency of our methodology.