| With the increasing of Internet information and the reducing of the user’s terminal size, establishing a uniform data center to provide users with customized remote storage service is the inevitable trend of current storage technology. As the revolutionary technologies of next generation in the field of IT, cloud storage technology has been the research focus in the field of cloud computing. In the past few years, domestic and foreign academia and industry have been spent a lot of manpower and material resources in researching and developing cloud storage technology and product. So that new technologies and products are emerging in an endless stream. Cloud storage technology relates to network storage technologies, virtualization technology, data transmission technology, distributed storage technology and so on, and has good academic research prospects and value.Cloud storage systems take advantage of storage virtualization technology and network storage technologies to provide virtual machine clients with original block-level storage device, allowing users installing file system or database on such block-level storage device according to their own needs. There are two disadvantages with existing block level storage system:1.some of them are tightly coupled with their cloud computing environments. As a result, it’s hard to extend them to support other cloud computing platforms;2.There is bottleneck of volume server in system, which seriously affects the performance and reliability of the whole system.In this paper we present and implement a virtualized block-level cloud storage system with multiple volume servers, under the premise of researching other block storage system.At first, we design a multiple volume servers’architecture and its workflows. Under this architecture, system manages the shared storage device with CLVM, manages client virtualmachine with Xen, provides customer’s virtualmachines with independent and permanent block-level storage through iSCSI, and coordinates each system module to work orderly. A set of experiments show that the design of multiple volume servers eliminates single point bottleneck of the system, improves the I/O throughputs for the system compared with single volume design.Then, we propose a fault detection processing mechanism, which supports volume servers’ Listen-Detect-Switch in its running time. Because of this mechanism, volume servers can be switched when they are breaking down, and won’t lead to the system service interrupt when some of them don’t work. So it ensures reliability of the system.After that, we present both static and dynamic load balance strategies between multiple volume servers by modeling load and performance of volume servers. Static load balance strategy choices the volume server with lightest work load to serve the Client VM, keeping the system away from load unbalanced state. Dynamic load balance strategy decides the best way to deploy iSCSI connections based on genetic algorithm. So that cloud storage system can balance the load of each volume servers according to the simulatedresult of genetic algorithm when the volume servers’ cluster is in a load unbalanced state. Extensive experimental results show that load balance strategies between multiply volume servers remarkably improve system I/O throughput rate and I/O times per second.Finally, we summarize our main work and contribution in researching and implementing block-level cloud system, and look forward to the future work of our system. |
PDF Full Text Request