We are interested in partitioning sparse rectangular matrices for parallel processing. The partitioning problem has been well-studied in the square symmetric case, but the rectangular problem has received very little attention. We will formalize the rectangular matrix partitioning problem and discuss several methods for solving it. We will extend the spectral partitioning method for symmetric matrices to the rectangular case and compare this method to three new methods — the alternating partitioning method and two hybrid methods. The hybrid methods will be shown to be best.
PDF and postscript files are preprints.
@inproceedings{Ko98,
author = {Tamara G. Kolda},
title = {Partitioning Sparse Rectangular Matrices for Parallel Processing},
booktitle = {Solving Irregularly Structured Problems in Parallel},
eventtitle = {5th International Symposium, IRREGULAR'98},
venue = {Berkeley, CA},
eventdate = {1998-08-09/1998-08-11},
editor = {A. Ferreira and others},
series = {Lecture Notes in Computer Science},
volume = {1457},
number = {1457},
publisher = {Springer Berlin Heidelberg},
pages = {68-79},
year = {1998},
doi = {10.1007/BFb0018528},
}