Research And Implementation Of A Device Driver Stack Of Virtual Storage Technology

Virtual storage centralized management of heterogeneous devices, reducing storage equipment investment, improve the efficiency of storage. The typical virtualization management solutions from the the angle of the host, device and network to provide some common solution.However, there are still a lot of specific problems to be solved, such as the parallel I/O scheduling; high bandwidth of the task conditions transmission bottlenecks; and the loss of the services of the "hot spots" data in the network application.These issues are difficult to be resolved through the common virtualization, and there is a need for in-depth study to the underlying operating system.The main work and results of this thesis are as follows:First, analysis the virtualization principle of the logical drive and interface driver of the device driver stack, designs and implements a virtual network disk array system, which improve the system’s overall I/O performance.Second, combining the advantages of virtual storage of the logical drive layer, to build a distributed storage on the network disk array system and parallel scheduling policy based on a distributed storage,implement the parallelization of transmission, reduce data transmission path.Third, in accordance with the hotspot data access of the virtual network disk array system led to the problem of load imbalance,build a dynamic copy data migration strategy in the the logical driver layer. Analyse the key technology of to obtain information on the load, determination of hotspot data, the number of copies of the management and the management of migration space.Test results show that the dynamic increase or decrease in data copy can solve the problems of the hotspot data load imbalance and the I/O bottlenecks of a single disk array.Fourth, in order to further optimize I/O performance of virtual disk array system, implement the merge request queue task (MQT) algorithm in the queuing thread of the interface driver layer, and set jump the queue variable for each request task.Compared with the traditional algorithm, MQT algorithm can eliminate the phenomenon of "hunger" for certain requests, reduce seek time and rotational delay significantly, improve system throughput and reduce the average response time of the request.Therefore, when the generic virtual storage scheme is not a good solution to the specific problems which are brought by the storage, achieve the virtualization storage in the underlying drive is a better choice.