The Study Of P2P Network Architecture Based Butterfly Network

With the continuous development of the Internet, information stored in computernetworks is becoming increasingly diverse and more importantly, massive. As aconsequence, the storage needs of desktop users are also continuously increasing,measuring from initially Mega bytes to nowadays Tera bytes. It is foreseeable that, inthe near future, the storage needs of desktop users will be on the basis of Peta bytes.Therefore, we need to find storage solutions that can satisfy this increasing demand.Based on the idea of distributed storage, this paper proposes an Internet-baseddistributed file storage architecture, which, together with desktop virtualization,forms a distributed virtual storage system.After analyzing the limitations of the current DHT algorithms, this paperproposes an SF-DHT network architecture based on butterfly networks, and designsand implements the SF-DHT-based P2P network algorithm. In SF-DHT, peers are notcomputers (or IP addresses) as in traditional P2P networks; instead, each nodeconsists of three components---the service component, data component, and storagecomponent, each of which serves as a peer in the SF-DHT based P2P networks.Based on the SF-DHT algorithm, this paper designs and implements a distributedstorage system---the SF-STORE system, including all the three components of a nodein an SF-DHT P2P network, namely, the service component, the data component, andthe storage component. In particular, in the service component, it supports virtual diskdata access, enabling users to access the SF-STORE system in the same way as toaccess local disks.Finally, this paper compares the performance of the SF-STORE system with thatof eMule and CIFS. This paper investigates the performance of both the SF-DHTnetwork and the SF-STORE distributed storage system. The performance of SF-DHTnetwork is measured in terms of positioning time, positioning accuracy, and routingdelay, and is compared with that of a well-known P2P storage system eMule. Theperformance of the SF-STORE system is measured in terms of both I/O andthroughput. The I/O performance is measured using the benchmark program IOZONE,and compared with that of CIFS. The throughput is measured using a test programwritten with the MPI programming API, and compared with that of eMule.