The Design And Implementation Of Distributed Storage Subsystem Based On The Two Level Mapping

Along with the construction of the next generation of broadband wireless mobilenetwork, the mobile control network has raised higher requirements in backstagestorage for the sake of meeting the demand of telecommunication. The need ofavailability and scalability that proposed by users can’t be satisfied by the traditionalstorage solutions, especially in the circumstance of mass users and data: the systemcan’t scale with the number of uses, and it also can’t handle the concurrent access todata.This thesis first introduces some popular distributed file systems around the world,and then describes the design and implementation of resource locating mechanism andthe data storage module in a distributed file system CSTORE.CSTORE adopted the pure distributed architecture, and it stores metadata and datain two different clusters. And with the recovery strategy and load balancing strategy, theresource locating mechanism that based on two level mapping can provide goodscalability. The separation between metadata and data decouples the client access tothem, makes it easy to optimize their management and performance respectively.Furthermore, the utilization of clusters promotes the distributed administration ofservers, avoiding the risk of single failure.The CSTORE resource locating mechanism constructs the relation betweenresources and servers. It provides clients with a reliable locating method. With the twolevel mapping, each node in system can calculate the location of resource independently.It not only takes advantage of clients, but also reduces the pressure of servers. Itspseudo-random property and the transform of the mapping helped in load balancing andsystem scaling.The data storage module provides users with data storing and accessing service.The data stored in the system will be divided into different resource unit according tothe resource locating mechanism. In order to exchange the sequence and push data, thecopies of the unit build heartbeat connections. Thus the data availability and consistencyare assured. As for the fault-tolerance, the system can tolerant a certain range of failure because of the redundant mechanism. And the recovery strategy makes data storagemodule serve for clients as soon as possible. All the blocks stored in module has aunique ID, accompany with it, deduplication is implemented. The advantage is obvious:the deduplication saves the cost on storage and administration, and improves theeffectiveness of disk space.In the last, the thesis tests the functionality and performance of the system, whichproves its validity and high efficiency.