Research On Similarity Search For High-dimensional Data Objects In P2P Networks

In order to achieve the access of massive multi-media information on P2P networks efficiently, the data's processing and retrieving have become a hotspot of current research. Traditional relational database and the precise query base on it can not meet the requirement of applications; content-based retrieval of multi-media data is making progress in research. Content-based information retrieval of multi-media, such as text, images, audio, video et al, needs to establish an effective indexing structure to support variety complex queries. This leads to a very challenging research topic: how to implement a similarity search system base on P2P for massive, high-dimensional data.The work carried out in this paper as follows:(1) Proposed an indexing method: PLCID (modified iDistance based on Proximity Location Code). When retrieving data, the technique uses a two-tier filtering process which greatly narrows searching scope and reduces the times of distance calculations between high-dimensional data. So it greatly improves the performance of data searching.(2) Based on the indexing structure mentioned above, we achieve a high-dimensional data retrieving system on a structure P2P networks. Experiments show that compared with the original iDistance indexing method, our technique gets some improvements in terms of time performance and system overhead.(3) We introduce a solution for load balance problem in our system. And for the core issue: how to measure the node's load overhead, propose a algorithm named LVDCB(Load Variety and Distribution Cost based algorithm) to solve the problem. Experimental study shows this method works well for the system's load balancing.