Data-base processing on a cube-connected multicomputer

Thesis/Dissertation · Thu Jan 01 04:00:00 EST 1987

OSTI ID:7106918

Frieder, O

A boolean cube-connected relational data-base machine is developed. Strategies for performing the basic relational data-base operations are presented. The strategies are termed dynamic and are novel in that they account for the nonuniform data distribution across parallel paths by redistributing the data as part of the data-base algorithms. Obtaining uniform data distributions is important in parallel architectures in order to attain close to optimal performance. The cube interconnection scheme employed subsumes many other interconnection structures such as the tree, ring, etc. Within the proposed architecture, an overall query processing approach using these algorithms and a flexible cube controlling a cube control structure are presented. The scheduling method used groups that needed data-base operations to reduce scheduling complexity, while still supporting intra- and inter-query concurrency and minimizing communication delay. Besides being novel by dynamically redistributing data resulting from intermediate calculations, this research demonstrates the possibility of implementing fast data-base operations on a general, multi-purpose multicomputer system with lower response times than special-purpose data-base architectures.

• Research Organization: Michigan Univ., Ann Arbor (USA)
• OSTI ID: 7106918
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990210* -- Supercomputers-- (1987-1989)
990300 -- Information Handling
ALGORITHMS
ARRAY PROCESSORS
COMPUTER ARCHITECTURE
DATA PROCESSING
DISTRIBUTED DATA PROCESSING
MATHEMATICAL LOGIC
PARALLEL PROCESSING
PROCESSING
PROGRAMMING
TASK SCHEDULING