面向QEMU的分布式块存储系统的设计与实现

Design and Implementation of Distributed Block Storage for QEMU

计算系统虚拟化是当前的主流趋势,然而为虚拟机磁盘提供存储的现有后端系统常直接沿袭分布式文件系统的设计,导致一系列虚拟磁盘I/O场景下不必要的开销。提出的COMET系统,是一个面向QEMU的分布式块存储系统。它基于Proactor模式,针对虚拟磁盘I/O特性扬弃文件系统的语义并在数据读写与一致性等方面采用精简合理的设计,尽可能缩短I/O路径,在具备线性扩展能力、高可用性的同时,实现了写时复制的磁盘快照和写时复制的磁盘克隆等创新功能。测试数据表明,COMET在部分I/O场景下拥有足以媲关乃至超越成熟开源系统的性能。

Virtualization of computing system has become a main trend.However,currentbackend storage implementationsfor virtual machine disks usuallyreference directly from distributed file system,resulting in a series of unnecessary overhead in virtual disk I/O scenarios.The COMET system presented by this paper is a distributed block storage systems for QEMU.Based on Proactor mode,COMET abandons the semantics of the file system,takes reasonable design in data access as well as consistency and maintains a minized I/O path for virtual disk I/O scenarios.Withlinear scalability,high availability,COMET implemented innovative features such as write-on-copy disk snapshot and write-on-copyclone.Results show that COMET has a performance comparable to or even betterthan mature open source systems in certain I/O scenarios.