【正文】 P2P， DHT， Chord，查找，尋路， IPv6，拓?fù)?，層次化，文件?享 中國科學(xué)技術(shù)大學(xué)碩士學(xué)位論文 Abstract III Abstract With the great improvement of PC performance and the fast growth of Inter users, there emerges a vast quantity of puting and storage resources on the Inter edge. P2P (peertopeer) technology can be an effective means to harness these resources, which accounts for the fact that P2P applications are being more and more popular these days. In a P2P system, all peers are identical regarding functionality. Unlike the traditional C/S (client/server) model, there are no dedicated servers and peers can directly municate with each other for data transmission. P2P can solve the problems of single point failure and performance bottle encountered by C/S model. A fundamental problem that confronts a largescale P2P system is the efficient location of the node that stores the desired date item. However, the first generation of P2P systems did not address the problem well. Napster has a centralized index server where scalability can be limited by the machine power and the work bandwidth of the central point. Gnutella employs a messaging mechanism that is based on flooding, which can impose heavy burden on works and thus promise its scalability. To address the problem, several research groups independently proposed DHT (distributed hash table) systems, which include Chord, CAN, Pastry and Tapestry. DHTs reanize peers into an overlay in the application level, distribute file indexes into the work, and route queries through the overlay. DHTs are robust in the face of failures, attacks and unexpectedly high loads. They are scalable, achieving large system sizes without incurring undue overhead. They are selfconfiguring, automatically incorporating new nodes without manual intervention or oversight. They provide a simple and flexible interface and are simultaneously usable by many applications. However, DHTs are still faced with many problems, one of which is the fact that most DHTs do not take into account physical work topology in their original design, thus resulting in high routing latency and low efficiency. Therefore, to improve routing performance is an important direction for research on DHTbased P2P. While centering on the issue of routing enhancement, the author has conducted an indepth research on how to extract topology information and how to utilize that information to construct topologyaware DHT systems. In Chapter 3, we propose three solutions, which are called DHT with hierarchical identifiers, embedded DHT and hierarchical DHT. To illustrate our solutions, we build Chord6, eChord and hChord all upon the original Chord system. Analysis and simulation results prove that our solutions can greatly improve routing efficiency in Chord. Currently, a new generation of applications has been proposed on top of DHTs. In this paper, we also design a widearea filesharing system based on Chord6, validating the effectiveness of our research work on DHT routing enhancement. The major contributions of this paper are listed as follows: 1. Propose a novel method to extract topology information from IPv6 address prefixes. We notice that IPv6 addresses are assigned in a hierarchical way so that nodes with the same prefix are in the same autonomous domain. Therefore peers in a DHT system can learn their location information from their own IPv6 addresses. 2. Devise a smart scheme to exploit the IPv6 address hierarchical feature, so as to construct an efficient version of Chord dubbed Chord6. We propose that node identifiers can be divided into several parts and thus be produced separately. For a node identifier divided 中國科學(xué)技術(shù)大學(xué)碩士學(xué)位論文 Abstract IV into two parts, the higher bits can be obtained by hashing the shared address prefix among all nodes within the same AS, and the lower bits are the hash result of the rest of the IPv6 address. As a result, topologically close peers shall also be adjacent in the overlay. An important advantage of our scheme is that it is very simple and barely modifies the original Chord. Simulation results have shown that our method can significantly reduce interdomain traffic that causes the long routing latency. 3. Devise a novel scheme to construct embedded DHT, which can not only improve the routing efficiency, but also inherit the loadbalancing feature of the original DHT. First, nodes independently insert their location information into DHT systems as they do with file indexes. Then, a newly joined node can utilize DHT to get a plete list of all nodes that are close to it in the underlying physical works. Finally, nodes within the same domains are anized into many local DHTs which are then embedded into a global DHT prised of all nodes. Thus, routing can be conducted in local DHTs first, and pass through each other (if necessary) with the aid of the global DHT, which means that interdomain traffic can be minimized to the extreme. To illustrate the feasibility and effectiveness of the scheme, we construct eChord upon the original Chord system. Analysis and simulation demonstrate that our scheme is very effective. 4. Propose a new kind of hierarchical DHT dubbed hChord, in which topologically close nodes are grouped in the overlay and keys are stored in a hierarchical way. Analysis show that hChord can isolate the effect of dynamic nodes within small groups for better scalability and stability, and show improved performance with partial queries. 5. Present a prototype design of an IPv6based widearea file sharing system based on Chord6. Keywords： P2P， DHT， Chord， look up, routing， IPv6， topology， hierarchical, file sharing 中國科學(xué)技術(shù)大學(xué)碩士學(xué)位論文 目錄 V 目錄 摘要 ...........