[sbaffini]

So, if I understand correctly the results, there is indeed a problem when running with 2 cores on randomly ordered grids.

Or, depending from how one can categorize the 1 core results here, the randomly ordered case is indeed more costy.

[flotus1]

Edit: I completely misinterpreted my own results. Of course there is a run time difference for all versions of Metis and ParMetis depending on locality of the graph
The most important result for me is that ParMetis when run in parallel, can not achieve the same level of quality as the serial version. From other tests, I know that the edgecut value can have a severe impact on solver performance. 5% difference in edgecut translated to 4% difference in solver performance.

[aerosayan]

I don't use parmetis, but if you need help on basic fortran-c interoperability, I created a few example codes for my own projects. I show how to exchange variables, structs, arrays, matrices, and other things between fortran and c.


LINK : https://github.com/aerosayan/fortran-c-interop

[flotus1]

Well there is only one real problem remaining

Copyright & License Notice
ParMETIS is copyrighted by the Regents of the University of Minnesota. It can be freely used for educational and research purposes by non-profit institutions and US government agencies only. Other organizations are allowed to use ParMETIS only for evaluation purposes, and any further uses will require prior approval.

I get the strong feeling that ParMetis is effectively abandonware. What's worse, I haven't gotten any response to my inquiry yet.
Got any other ideas how to get in touch with the copyright holders?

[sbaffini]

Originally Posted by flotus1

Well there is only one real problem remaining

I get the strong feeling that ParMetis is effectively abandonware. What's worse, I haven't gotten any response to my inquiry yet.
Got any other ideas how to get in touch with the copyright holders?

https://github.com/KarypisLab/ParMETIS

But I feel that this isn't much different from a direct inquiry