Break of parallel calculation on 2/3 or more nodes

[Yuan Yin]

Dear developers,

Luckily, I can run a full CSC calculation on any single node on my servers. But a problem with the parallel calculation of DMFTwDFT codes on 2 or more nodes (e.g., node1: 20cores +node2: 20cores, equal to 40 cores) happens. Usually, the first 3-5 calculation loops run well, then the programs (dmft.x or ctqmc) becomes zombie. Everything stopped.

Here is the state of programs during running:

Could you help me to figure out what happens during the program execution?

Best thanks!

Yuan

[Uthpala Herath]

Dear Yuan, 

I'm glad you got the calculations to run. 

Regarding the multiple node issue, I feel it has to do something with how the interconnects are set up on your cluster. 

Can you try to see if the issue persists with other programs?

Also, your system administrator might be able to help out too. 

Thank you, 

Best,

Uthpala

--
You received this message because you are subscribed to the Google Groups "DMFTwDFT" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dmftwdft+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/dmftwdft/c99ae4cd-fd2e-4295-bee8-de5eb16378f2n%40googlegroups.com.

--

Uthpala Herath

Postdoctoral Associate

Department of Mechanical Engineering and Materials Science

Duke University

Durham, NC

Tel. (304) 216-2535

Email: uthpala...@duke.edu

Website: https://uthpalaherath.github.io/