The Study Of Dynamic Heterogeneous Virtual Disks Array And Its Pivotal Technologies

With advancement of micro-electronic technology, the speed of memory and the performance of microprocessor are improved rapidly, while the performance of storage system is very low because of its mechanical movement limitation. In order to alleviate I/O bottleneck problem and acquire better Cost/Performance I/O service, people has done lots of studies on Virtual Disk Array in cluster computering system. Our primary motivation is to organize the more and more distributed and heterogeneous I/O resource and optimize the organization. Main work of the dissertation is as followings:Above all, DHVDAS, Dynamic Heterogeneous Virtual Diaks Array System, is presented to construct heterogeneous and distributed I/O resource into a parallel, single system image and virtual disks array. DHVDAS can establish capabilities of high effective, error tolerance, transparency to user and etc.Then, we have presented the dynamic disks partition strategy under the instruction of OCD, Optimal Concurrency Degree of I/O reqquest. First we have done detail study of RFBO method to decide OCD (Optimal Concurrency Degree) of request. Then according to request OCD, we provide the algorithm of Maximum satisfaction and Minimum satisfaction strategy for the reorganization of system disks . By using these strategies, disks partition can satisfy I/O request concurrency need adaptively.Third, we have presented NLZP method to select OSUS, Optimal Stripe Unit Size, for data placement in HDG, Heterogeneous Disks Group. One popular type of disk group interleaves data in block-sized units, which is called data striping. And SUS, Stripe Unit Size, is one of the most crucial elements in data striping. This paper investigates in OSUS of HDG, and brings forward the method for OSUS choosing according to four factors: N (Number of disks of HDG), L (average queue Length of HDG), Z (Average Request Size), and P (different Product characteristic of each device). Disk group can make full use of each heterogeneous disk performance with NLZP OSUS selection.Finally, we present some algorithms for system dynamic reorganization. DHVDASis a dynamic system. Each change will bring with much extra overhead. In order to decrease its influence: first a TDO method for reorganization decision is brought forward. Then in order to decrease the datum having to be moved because of reorganization, the dissertation presents D/H mapping algorithm. And an online reorganization strategy named PBR is put forward. PBR strategy is a better trade-off between the system reorganization time and service availability.