linux - How to partition the graph with edge weights using METIS such that the edgecut is least? -
i have metis input file edge weights of graph. , want partition graph metis, such edgecut in metis summary report least possible metis. can rb, kway or other algorithms or options.
so options work best?
at moment, following options have worked best me.
gpmetis -ptype=rb metis.input.file num.of.partitions
using k-way partitioning (-ptype=kway) let's choose minimize edgecut (-objtype=cut) or minimize total communications volume (-objtype=vol). these 2 concepts similar (refer metis manual).
i found minimizing total communication volume worked better edgecut, because, when edges had weights, edgecut in metis defined sum of edge weights cut. minimizing "edgecut" turned out produce more border vertices.
edgecut did perform lot better if edge weights (temporarily) set 1, it's simpler use other option.
source: metis manual , personal experience.
Comments
Post a Comment