面向多核环境的BDD工具设计与优化
首发时间:2017-05-02
摘要:针对传统二分决策图(BDD)可扩展性(Scalability)问题,提出了一种面向多核处理器的并行BDD库的方法,利用顺序锁(seqlock)以及原子性比较并交换指令(CAS)优化了并发计算表(Computed Table)与唯一表(Unique Table),避免了在读取计算表与唯一表时的锁操作开销,有效地提高了系统的可扩展性。实验结果表明并行二分决策图库能够在64核服务器上相比目前最好的系统BuDDy达到高达25倍的性能提升,自身可扩展性高达30倍。
For information in English, please click here
Design and implementation of a parallel BDD package
Abstract:A parallel BDD algorithm is proposed to solve the problem of poor scalability of traditional BDD package. Seqlock and compare-and-swap are used to avoid acquiring locks during reading Computed Table and Unique Table. The experimental result shows parallel BDD algorithm can outperforms the state-of-the-art BDD package (namely BuDDy) by 25X speedup and it has good scalability (30X) on a 64 cores server.
Keywords: Parallel Computation Multicore BDD Scalability
论文图表:
引用
No.4729032119672014****
同行评议
共计0人参与
勘误表
面向多核环境的BDD工具设计与优化
评论
全部评论0/1000