一种基于副本机制的智能存储系统I/O优化方法

Optimization method for I/O of IND storage system based on replica mechanism

针对智能存储系统(IND)I/O部件级性能难以得到提升的问题,在分析I/O请求过程基础上设计了一种新的副本机制.使用可编程NIC把只读高频访问热点数据的副本存储在NIC的副本空间,在IND单元内存中建立一个副本元数据表以指引IND单元对副本空间进行管理,设计了在线副本候选算法(DCA),以确保NIC空间副本数据的时效性.读请求时副本数据不再通过总线传送,数据包到达NIC缓冲区时把副本数据填充在相应的位置再发送出去,可以减少总线的流量,进而提高系统的吞吐率和现有设备的性能.试验结果表明,该方法降低IND总线流量30%以上,提高系统吞吐率40%以上,显著提高了系统性能,具有较好的理论与应用价值.

To solve the problem that the performance of I/O components for IND storage system is hard to be improved, a novel replica mechanism was designed through analyzing the I/O request process. The programmable NIC was used to store the the replica of read-only high-frequency accessing hot data in the replica space of NIC. A replica metadata table was established in the memory of IND units to guide the management of IND units to the replica space. The online DCA was designed to ensure the timeliness of replica data in NIC space. In the requst process, the replica data would not be transferred by bus. When the data packet reached the INC buffer zone, the replica data would be filled in the corresponding positions and then sent out, which could reduce the flow of the bus and further enhance the throughput of the system as well as improve the performance of the existing devices. The experimental results show that the proposed method can reduce the flow of the IND bus by above 30%, increase the throughput of the system by above 40%, improve the performane of the system obviously, and possesses the better theoretical and practical values.