| With the development of peer-to-peer computing paradigm, network bandwithand, the capacity of Internet-based systems, the end-systems ignored in the past become a valued computing resource. How to utilize the end-systems resources to build a distributed storage system with large-scale, high scalability, reliability and performance in dynamic peer-to-peer network environment is a hot research topic in recently years.In this thesis, firstly, it discusses and summarizes the development and actuality of distributed storage systems. Based on systematically analyzing and summarizing the relevant works on peer-to-peer technology, it designs a P2P-based distributed storage system: PPDSS. Baesd on a highly scalable P2P-based architecture, PPDSS organizes large numbers of nodes distributed in Internet into a united overlay network. Due to the complexity of network environment, it is inevitable for the problem of single-stored node failure. Hence, the mechanism of multi-replica has been introduced into the system, which is in order to evaluate the usability and reliability. But while the management of the replica, two significant issues, the redundancy and consistency of the replica, should be handled first.Among the existing techniques, it is universally accepted that the management of replica is based on the frequency. The more frequent users access a file, the more replicas this file enjoys. However, at the same time, this solution cannot actually solve real-world problem. Assuming that the capacity of the sever is so adequate to handle the concurrent easily access from the client, then there is no need to increase the replica to improve the user experience.From the analysis above, it reveals that there is no necessary relationship between the frequency of access and the number of replica. That's why this paper tries to provide a better mechanism of replica management, UERM, which is totally based on the feature of P2P proxy used by PPDSS. For instance, when the system decides to increase replica completely depends on a value named User Experience, which is reled to all the influential factors such as the number of downloading processes, the priority of each process, the average downloading speed in a unit time, the total bandwithand and so on. Only when this User Experience is lower than a standard value can the number of replica be increased. The major advantage of UERM is to dynamically manage the replica without any burden or overhead introduced into the system.Considerable as the benefits that the replica-redundancy advances the usability of data and makes it possible for the system to tolerate fault are, at the same time, the replica-consistent issue has been brought about. Considering that PPDSS is such a storage system that would have more-read but less-write operation, a new consistent solution for this PPDSS model has been proposed. The solution involves the version management toward the files or blocks, and the mechanism of two-phase commit applied to control all the modifications of resources in the system.By passing the functional and performance test for whole system, the result demonstrates that this system can not only enhance the usability of data, decrease the delay of access the resources, but also save the bandwithand for the data consistent. Better storage service can be provided for clients by these benefits. |
PDF Full Text Request