摘要： 针对当前只能在一个集成光电芯片上实现容量为几十个分组的光先来先服务(FCFS)缓存，大规模的光缓存仍无法实现的问题，利用接入链路速率远小于骨干链路速率这一特性，在牺牲25%的链路利用率的条件下，讨论小缓存能否满足采用CIOQ缓存队列的全光分组交换路由器的性能需求。从理论上证明采用CIOQ缓存队列的路由器的缓存需求小于采用OQ缓存队列的缓存需求，通过分析和仿真发现，20个分组的缓存即可满性能需求。

关键词: 缓存, 分组交换, 路由器

Abstract: An integrate photoelectric chip just realizes small light First Come First Serve(FCFS) buffer, and can not achieve big buffer. This paper makes use of the characteristic that access link rate is highly less than bankbone link bandwidth, and under the condition of sacrificing 25% of bankbone link bandwidth, discusses if small buffer can suffice the performance requirement of an all-optical packet switching router based on CIOQ buffer queue. It demonstrates the buffer requirement of router based on CIOQ queue is highly less than the buffer requirement of router based on OQ buffer queue. Through analysis and emulation, it finds 20 packets can suffice performance requirement of bankbone router.

Key words: buffer, packet switching, router

中图分类号: 

• TP393