search

OrderN / CONQUEST-release

Full public release of large scale and linear scaling DFT code CONQUEST
http://www.order-n.org/
MIT License
96 stars 25 forks source link

Improve error message from partitioning when there are more processes than atoms #106

Closed Euicheol-Shin closed 1 year ago

Euicheol-Shin commented 2 years ago

Many thanks to the Conquest developer team for their hard work and open source. Recently, after installing the code, I keep running into an error while doing exercises for Silicon.

Each node I use has 12 cores. When using the silicon example with 12 cores of a single node, the following error message appears,

Error in process 1 sfc_partitions_module: too few occupied partitions for number of processors 8 12

If one node uses 8 cores, it runs successfully. Since it is still in the early stages of public disclosure, there are no detailed forums or other discussions about the content, so I leave a problem here.

Thank you, Euicheol

davidbowler commented 2 years ago

Hi Euicheol. Thanks for your interest in Conquest.

We require at least one atom per process to run Conquest, so with the simple 8 atom silicon cell, you cannot run on more than 8 processes (or cores/processors if you prefer).

We should make the error clearer - I will change the name of the issue to reflect this.

I hope that helps.

Dave

tsuyoshi38 commented 2 years ago

Thank you very much for your feedback. How many atoms do you have in your system ?
If it only includes 8 atoms, CONQUEST now assumes that each MPI process should include at least one atom, for parallel efficiency. Thus, the code runs with 8 processes, but not with 12 processes.

For the forum, we have a mailing list. (but, I forgot its name..)

Euicheol-Shin commented 2 years ago

Hi Euicheol. Thanks for your interest in Conquest.

We require at least one atom per process to run Conquest, so with the simple 8 atom silicon cell, you cannot run on more than 8 processes (or cores/processors if you prefer).

We should make the error clearer - I will change the name of the issue to reflect this.

I hope that helps.

Dave

Thanks for the very quick feedback.

I have a clear understanding of the problem. Thank you for your hard work.

Euicheol

Euicheol-Shin commented 2 years ago

Thank you very much for your feedback. How many atoms do you have in your system ? If it only includes 8 atoms, CONQUEST now assumes that each MPI process should include at least one atom, for parallel efficiency. Thus, the code runs with 8 processes, but not with 12 processes.

For the forum, we have a mailing list. (but, I forgot its name..)

Thank you, Tsuyoshi

If you can give the mailing list of the forum, it would be more helpful.

Euicheol