Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/12534
Authors: Baburao, Kanase Patil Padmaja
Issue Date: 2010
Abstract: With the growing use of search engines, information retrieval has become a latest research area. The growth of internet and explosion of information are the major factors due to which the size of documents has increased many folds. Thus there is a need of effective and efficient information retrieval mechanism. Latent Semantic Indexing (LSI) is one of the techniques for information retrieval which is very effective in correlating and retrieving relevant documents. The critical step involved in LSI algorithm is Singular Value Decomposition (SVD). SVD is a mathematical technique which is basically a matrix decomposition method. The SVD is highly effective to derive the semantic relevance, but it is computationally very expensive in terms of time and memory. Thus SVD becomes a bottleneck for quick retrieval of matching documents from large database. This necessitates the optimization of algorithm by parallelization using high performing architecture. Multicore processors can be used for such type of problems which are computationally intensive. The Cell Broadband Engine is one such multicore processor consisting of a traditional PowerPC based master core meant to run the operating system, and 8 delegate slave processors built for compute intensive processing. This work introduces a modification on the serial singular value decomposition algorithm. It describes parallel implementation of the modified algorithm on Cell BE and issues involved. Exposure of system level optimization features in Cell BE has been employed on algorithm specific operations to achieve improvements to a great extent. The implementation achieves significant performance, thereby giving about 8 times speedup over sequential implementation.
Other Identifiers: M.Tech
Research Supervisor/ Guide: Singh, Kuldip
Mittal, Ankush
metadata.dc.type: M.Tech Dessertation
Appears in Collections:MASTERS' DISSERTATIONS (E & C)

Files in This Item:
File Description SizeFormat 
ECDG24043.pdf1.6 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.