A flexible parallel runtime for large scale block-based matrix multiplication

Authors:
Keyan Liu;Shaohua Song;Ningnan Zhou;Yanyu Ma
Affiliations:
Nokia-Simense-Network CTO Research, Beijing, P.R.C;RenMin University of China, Beijing, P.R.C;RenMin University of China, Beijing, P.R.C;Nokia-Simense-Network CTO Research, Beijing, P.R.C
Venue:
APWeb'12 Proceedings of the 14th international conference on Web Technologies and Applications
Year:
2012

Citing 2
Cited 0

MapReduce: simplified data processing on large clusters

OSDI'04 Proceedings of the 6th conference on Symposium on Opearting Systems Design & Implementation - Volume 6
HAMA: An Efficient Matrix Computation with the MapReduce Framework

CLOUDCOM '10 Proceedings of the 2010 IEEE Second International Conference on Cloud Computing Technology and Science

Quantified Score

Hi-index	0.00

Visualization

Abstract

Block-based matrix multiplication plays an important role in statics computing. It is hard to make large scale matrix multiplication in data statistics and analysis. A flexible parallel runtime for large scale block-based matrix is proposed in this paper. With MapReduce framework, four parallel matrix multiplication methods have been discussed. Three methods use the HDFS to be the storage and one method utilizes the Cloud storage to be the storage. The parallel runtime will determine to use the appropriate block-based matrix multiplication. Experiments have been made to test the proposed flexible parallel runtime with large scale randomly generated data and public matrix collection. The results have shown that the proposed runtime has a good effect to select the best matrix multiplication strategy.