Why the most overweight partition increases when the number of partition increases?

Questions, issues, or suggestions related to post-processing of model output.
Post Reply
jhuang54
Posts: 1
Joined: Wed May 15, 2019 3:01 pm

Why the most overweight partition increases when the number of partition increases?

Post by jhuang54 » Wed May 15, 2019 3:35 pm

Hi,

I am using gpmetis, a stand-alone program of METIS. The fmt parameter is 010 (set vertex weights for different vertex). I type the following command.

gpmetis -contig -seed x 8500NoSwRY npart

where x is seed value, npart is the number of parts that the graph will be partitioned into. For each npart, I used seed values from -900 to 900.


When I increases the number of partitions from 89 to 122, the most overweight partition increases. I don't know why.

I have attached the input file and the plot showing "number of partitions" vs "the most overweight partition". The vertical axis is the most overweight partition.

Thanks,
Jianqiao Huang

mgduda
Posts: 362
Joined: Mon Feb 26, 2018 7:35 pm

Re: Why the most overweight partition increases when the number of partition increases?

Post by mgduda » Wed May 29, 2019 8:19 pm

It looks like the attachment might not have gone through. In any case, I suspect that there might not be many MPAS users who are familiar enough with the details of Metis (which is not developed or supported by either NCAR or the Department of Energy) to post a helpful reply here. It may be better to see if you can get in contact with someone from the Karypis lab where Metis is developed.
NCAR/MMM

Post Reply

Return to “Post-processing”