Design And Implementation Of A Distributed Block Storage System For Virtual Machines In The Iaas Cloud

It’s believed that cloud computing is another great change since previous migration from mainframe to C/S architecture. Cloud computing covers three layers of serveries, which are SaaS(Software as a Service), PaaS(Platform as a Service) and IaaS(Infrastructure as a Service). Virtual machines is one of the core technologies in the cloud computing. And I/O subsystem becomes the bottleneck of the whole system, especially the virtual disk.Traditional storages such as DAS, NAS, SAN and distributed file system are hard to meet all demands in the IaaS. Neither capacity, performance, availability, scalability nor price can be sacrificed. Designing a distributed block storage system which provides a view of SAN brings us the simplification of the system architecture, the reduction of the file system layer overhead, higher performance and availablity, almost unlimited scalability, as well as low price. I believe it’s an important innovation.By combining a local data volume, a remote data volume and a journal volume creatively, the system provides almost as much availability as the traditional 3-replica way, while reducing capacity usage. Performance is improved by writing journal and local data volume concurrently, and updating remote data volume in the background. Data transmission also reduces network overhead. The test result shows that the system provides all the IOPS and most of the bandwidth which the lower layer device supports, and failure tolerance and recovery allows the I/O service going on with an acceptable performance without being interrupted.Several problems are left to be solved in the future. The network traffic of data transmission may lead to network congestion and further cascading failure, and the system should provide mechinism to control this. Hardware failure may damage part of the data volume silently, and the system should integrate a checksum method. The system perform a far-from-satisfactory small chunks sequential writing, and need furtuer optimization. The system can’t sense the existance of SSD, let alone taking advantage of that. This is an aspect to be improved.