面向NVMeoF存储网络的负载感知传输机制研究  

Research on a Load-Aware Transmission Mechanism for NVeoF Storage Network

作　　者：齐勇军[1] 汤海林[1] QI Yongjun;TANG Hailin(Guangdong Baiyun University Faculty of Megadata and Computer,510450,Guangzhou,PRC)

机构地区：[1]广东白云学院大数据与计算机学院,广州510450

出　　处：《江西科学》2022年第4期762-767,806,共7页Jiangxi Science

基　　金：广东省普通高校特色创新类项目(2020KTSCX163);广东省普通高校重点领域专项(2020ZDZX3009);广东省普通高校特色创新类项目(2018KTSCX256);广东白云学院重点科研项目(2019BYKYZ02)。

摘　　要：NVMe over TCP是构建大规模高性能存储系统的关键技术,是基于现有的数据中心网络基础设施和标准的TCP/IP软件协议,实现了NVMeoF(NVMe over Fabrics)存储网络。本文设计并实现了Load-Aware NVMeoF的消息处理机制LANoT(Load-Aware NVMe over TCP)。首先,采用基于聚合PDU的中断合并技术,缓解中断风暴问题,实现高吞吐量;其次,可以根据不同专用队列的I/O特性,匹配特殊的消息处理机制,有效提高应用的关键性能指标。在Linux内核中实现了LANoT原型系统。通过性能测试结果表明:与标准Linux内核中的NVMe over TCP实现相比,LANoT可以减少50%以上的CPU资源消耗,提高2倍以上的IOPS。NVMe over TCP is a key technology for building large-scale high-performance storage systems.It is based on the existing data center network infrastructure and standard TCP/IP software protocols to implement NVMeoF(NVMe over Fabrics)storage network.This article designs and implements Load-Aware NVMeoF's message processing mechanism LANoT(Load-Aware NVMe over TCP).First of all,it adopts the interrupt merging technology based on aggregated PDU to alleviate the interruption storm problem and achieve high throughput.Second,according to the I/O characteristics of different dedicated queues,special message processing mechanisms can be matched to effectively improve the key performance indicators of the application.This paper implements the LANoT prototype system in the Linux kernel.The performance test results show that:Compared with the implementation of NVMe over TCP in the standard Linux kernel,LANoT can reduce CPU resource consumption by more than 50%and increase IOPS by more than 2 times.

关 键 词：NVMe over Fabrics 消息处理机制 NVMe over TCP 存储网络 负载感知 

分 类 号：TP333[自动化与计算机技术—计算机系统结构]