摘要
提出了一种FC-AE-RDMA(光纤通道-航空电子环境-远程直接内存访问)协议的实现方法,采用内存区域作为目标端设备以缩短I/O访问路径,同时取消写操作中的流量控制信息单元以减少延迟时间.根据FC-AE-RDMA协议所描述的特征,为采用现场可编程门阵列实现的定制光纤通道控制硬件设计了支持RDMA传输的发起端和目标端驱动程序,并进行了光纤通道RDMA和FCP I/O访问的对比测试.测试结果显示:光纤通道RDMA比FCP具有更短的延迟,当执行最小的I/O请求(512byte)时,光纤通道RDMA的I/O读请求延迟时间为FCP I/O读请求延迟时间的49%;光纤通道RDMA的I/O写请求延迟时间为FCP I/O读请求延迟时间的29%,其中取消流量控制信息单元令使得请求延迟时间减少22%.
Practical approach was presented to implement FC AE-RDMA (fibre channel-avionics envi- ronment-remote direct memory access) protocol. This method used memory region as target device to shorten the I/O access path and disabled the flow control message in write operations to reduce latency. According to the characteristics described in the FC-AE-RDMA protocol, the initiator and target device drivers were designed to support RDMA (remote direct memory access) transmission on the costumed FC (fibre channel) controller implemented with FPGA (field programmable gate array). The I/O access performance of RDMA and FCP (fibre channel protocol) for SCSI was compared. The results show that FC-AE-RDMA outperforms FCP regarding I/O latency. On processing the mini- mum I/O request (512 byte), the read I/O latency for FC RDMA is 49% of that for FCP; while the write I/O latency for FC RDMA is 29% of that for FCP, and the removal of flow control message contributes 22% of the reduction.
出处
《华中科技大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2011年第6期71-74,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家高技术研究发展计划资助项目(2009AA01A402)
武汉光电国家实验室创新基金资助项目(Z080006)