光互连与电互连的并行运算性能比较  

Comparison of Parallel Computing Performance Between the Optical and Electrical Interconnects

作　　者：李倩[1] 乔耀军[1] 纪越峰[1] 

机构地区：[1]北京邮电大学信息光子学与光通信教育部重点实验室,北京100876

出　　处：《光学学报》2010年第2期498-502,共5页Acta Optica Sinica

基　　金：国家863计划(2007AA01Z2a6);国家973计划(2007CB310705);国家自然科学基金(60711140087);教育部博士点基金(200800130001);教育部创新团队项目(IRT0609);科技部国际合作计划(2006DFA11040)资助课题

摘　　要：光互连在短距离传输中的应用是未来大规模计算领域的关键技术。针对并行运算系统对高速高效传输的应用需求,研究了不同结构和不同传输方式下并行运算系统的性能,并从并行运算性能这一角度比较了光互连系统与电互连系统的优劣。采用网孔结构和超立方结构两种模型,用加速比和效率两个指标来评价光互连系统和电互连系统在并行运算方面的性能差异。得到光互连系统及电互连系统在进行并行计算时的加速比和效率与计算规模之间的关系,分析了计算规模对计算速比和效率的影响。通过加速比和效率这两个量的比较得到结论,在大规模并行计算中,光互连系统有着电互连系统不可比拟的优势。Optical interconnects for short-distance transmission are becoming more and more important along with the increasing demand on large-scale computing tasks. An analytic method to evaluate the computing performance of optical and electrical chip-to-chip interconnect systems from the parallel computing performance view is proposed. The mesh architecture model and hypercube architecture model are both investigated. The speedup and efficiency are selected to represent the parallel computing performance of an interconnect system. The relation between speedup and computing number,together with the relation between efficiency and computing number,are summarized and analyzed. The comparison shows that the parallel system based on optical chip-to-chip interconnects system has much higher speedup and efficiency than that based on electrical system.

关 键 词：光通信 光互连 并行体系结构 快速傅里叶变换 

分 类 号：O439[机械工程—光学工程]