Design And Implementation Of A QEMU Oriented Distributed Block Storage System

Computing system virtualization is the main trend in the present.The COMET distributed block storage system developed in this thesis is aimed at providing efficient and reliable logical disk for the virtual machine server in IaaS cloud services.Currently,KVM-based QEMU is a widely used virtual machine management software in production environment.As far as the known technology is concerned,the QEMU's existing back-end storage includes traditional storages such as SAN,NAS,DAS and distributed file systems such as Ceph,GlusterFS,and Sheepdog.However,the direct use of traditional storage is difficult to meet the scalability and performance requirements of the cloud environment,while the existing distributed systems have many deficiencies in the architecture.For instance,the core of Ceph concerns so much about the versatility and performance of Ceph-based virtual machine disk is poor.GlusterFS takes the weak consistency which is simple but not able to meet the high reliability requirements.Sheepdog introduced Gateway who extends the I/O path,further more,its multi thread simulation of asynchronous I/O cannot make full use of physical disk...A new IaaS virtual machine backend storage system,which can overcome these shortcomings,is valuable for both academic and industrial circles.This is the significance of developing COMET system.The COMET system developed by the author of the paper is a distributed system that provides block level storage for QEMU virtual machines in IaaS.Related work inlucdes:1)Design and implement the underlying Libcomet I/O library.2)Design the lightweight-centered distributed architecture with linear expansion,and implement the corresponding distributed data locating method.3)Design and implement a strong consistency algorithm for I/O of the virtual machine disk.4)Based on Paxos algorithm,implement the consistency of the event sequence among Topology Manager nodes.5)Implement both server-level and disk-level fault monitoring mechanism.6)Implement copy-on-write virtual machine disk snapshot and virtual machine disk cloning.7)Implement client of COMET system,which is embedded in QEMU code.COMET has good scalability,reliability and availability.Tests show that COMET can provide a better performance than the QEMU virtual machine dedicated storage system Sheepdog in most virtual machine disk I/O scenarios.