Show simple item record

dc.rights.licenseCC-BY-NC-ND
dc.contributor.advisorBisseling, R. H.
dc.contributor.authorOort, M. van
dc.date.accessioned2017-06-22T17:03:19Z
dc.date.available2017-06-22T17:03:19Z
dc.date.issued2017
dc.identifier.urihttps://studenttheses.uu.nl/handle/20.500.12932/25968
dc.description.abstractWe consider a number of modifications to the Mondriaan package for sparse matrix partitioning, with as main goal to improve run times of the software. Two new algorithms are considered, one of which aims at reducing load imbalance to enable Mondriaan to find partitionings with less communication volume, while the other aims at finding solutions with zero communication volume more quickly than Mondriaan currently does. The other modifications include an improved sorting function, an improvement in a data structure used in a core component of Mondriaan, and two modifications in the PGA matching algorithm implementation used in Mondriaan. We compare the performance of each modification against the previous version, Mondriaan 4.1, and we will discuss which modifications are suitable to include into a new version, Mondriaan 4.2. Experiments with a candidate version for Mondriaan 4.2 show a mean run time improvement of 25% compared to Mondriaan 4.1, while also significantly reducing the mean load imbalance and slightly reducing the mean communication volume.
dc.description.sponsorshipUtrecht University
dc.format.extent1149778
dc.format.mimetypeapplication/pdf
dc.language.isoen
dc.titleAccelerating the Mondriaan sparse matrix partitioning package
dc.type.contentMaster Thesis
dc.rights.accessrightsOpen Access
dc.subject.keywordsMondriaan,sparse matrix,partitioning
dc.subject.courseuuMathematical Sciences


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record