以状态子集为中心的并行模型检测算法-计算机系统应用.PDF

2016 年 第 25 卷 第 10 期 计 算 机 系 统 应 用 ① 以状态子集为中心的并行模型检测算法 张营飞 1,2, 谢 淼 1,2, 张 珩 1,2, 杨秋松 1 1( 中国科学院软件研究所 基础软件国家工程研究中心, 北京 100190) 2( 中国科学院大学, 北京 100190) 摘 要: 以线性时序逻辑LTL(Linear Temporal Logic)模型检测算法为研究对象, 提出以状态子集为中心的并行模 型检测算法. 针对传统单机多核算法同步开销大的缺点, 新算法充分利用状态子集的稠密特性动态调度任务, 从 而降低同步开销, 提高算法并行度. 本文基于轻量级单机图计算框架 Ligra, 结合检测过程中状态子集的特性, 设 计并实现新的在线(on-the-fly)模型检测算法. 与现有算法相比, 在模型检测的效率上可以提升 20-30%, 具有高扩 展性特征. 关键词: LTL 模型检测; 状态子集; 并行; 在线方法; 图计算 Parallel Model Checking Algorithm Based on State Subset ZHANG Ying-Fei1,2, XIE Miao1,2, ZHANG Heng1,2,YANG Qiu-Song1 1(Institute of Software, Chinese Academy of Sciences, Beijing 100190, China) 2(University of Chinese Academy of Sciences, Beijing 100190, China) Abstract: Logic model checking raise unique challenges to the efficiency and scalability of various algorithms. In this paper, it argues that the linear temporal logic model checking algorithm calls for an efficient and common solution. We then introduces a novel parallel on-the-fly model checking algorithm based on subset of states, namely SCPLM. According to the density property of subgraph with active states subset, SCPLM schedules parallel computing tasks dynamically, which has been proven to reduce the synchronization overhead significantly. It implements SCPLM on a lightweight parallel graph processing system, Ligra. A detailed evaluation using benchmarks shows that SCPLM outperforms the existing state-of-art algorithm by up to 1.2~1.3X, yet with much better scalability. Key words: LTL model checking; Subset of states; parallel; on-the-fly method; graph processing 1 引言