Gossip-based Resource Discovery Algorithms in Distributed Networks

• Main Authors: Gary Chung, 鍾嘉禮
• Other Authors: Yuh-Jzer Joung
• Format: Others
• Language: en_US
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/83612727385961838452

碩士 === 國立臺灣大學 === 資訊管理研究所 === 91 === Resource discovery problem [5] was first introduced by Harchol-Balter, Leighton and Lewin. When they are working on the problem of the web caching in large distributed system, they developed a randomized algorithm [5] to provide each node with “knowledge” about the topology of the network. A deterministic algorithm[6] with an O(log N) time complexity is presented by Kutten, Peleg and Vishkin later on. For resource update problem, Kempe, Kleinberg and Demers introduced a spatial gossip algorithm[7] that guarantees any new information generated will spread to nodes at distance d, with a very high probability, in O(log1+ξd) time steps. The algorithm has some good properties on exchanging information between nodes in a uniform density dynamic environment. Despite of the above efforts, in many cases, such as caching system, document system, or peer-to-peer file sharing system, we would like to access a nearby node faster than a node faraway. The algorithm proposed by Harchol-Balter, Leighton and Lewin, and the deterministic-form presented by Kutten, Peleg and Vishkin did not mention about locality in their results. Secondly, there is a single point of failure in the deterministic algorithm[6]. Due to the use of a star topology, the algorithm stops when it becomes a star. If the root node fails at that time, the whole topology information will be lost and there is no easy way to recover the graph by partial data. In order to overcome the problems mentioned, designing a suitable algorithm is needed. In this thesis, we will propose a resource discovery algorithm that considers both locality and performance.