earticle

영어

Distributed Hash Tables (DHT) provide a fault-tolerant and scalable means to store data blocks in P2P systems. In this thesis, we have proposed an improved version of CAN called “small-world DHT”. The key idea is to link each CAN node to a constant number of long-distance contacts, which are chosen according to the probability density function. This enables our scheme to achieve O(log2n) routing path length with O(1) routing table size per node. We have shown that the hybrid infrastructure of a structured overlay network (CAN) and a random graph (small-world model) not only preserve CAN’s simplicity, but also achieves a resilient and efficient DHT routing algorithm. We have also exploited the LookAhead-GREEDY routing algorithm, whereby each node obtains information about its neighbors’ neighbors by periodically exchanging the routing table entries with its neighbors. The routing algorithm improves the routing efficiency and enhances the network’s resilience to failure. Our scheme requires knowledge of the current network size in order to construct the long-distance links. To estimate the network size, the latter maintains a distributed binary partition tree and measures the size of the sample area by counting the number of leaves in the corresponding branch.