Yambo Community Forum

Dear All,

I am running a GW calculation for the bands of a bulk material. I wish to know how to optimize the memory usage on my cluster, but I was not able to find information on Yambo website.

For example, what the PAR_def_mode does? In the input file, it says "balanced, memory, workload", so I assume "memory" distributes the memory usage evenly to all processes, while workload does the same for the load. Am I right?

Best regards,
Roberto

Ciao Roberto,
you can find some documentation on parallelization strategies:
http://www.yambo-code.org/wiki/index.ph ... strategies
As a general rule, in order to distribute memory among cpus, the best option is to parallelize along bands (c,v).

If something it is not clear, or you need further clarifications do not hesitate to post in the forum.

Bests,

Daniele

Dear Roberto,
in case you do not want to care about yambo parallelization input variables.

Your interpretation is correct. The "memory" strategy should maximize the memory distribution.
However it is a recent variable and not very much tested, since all yambo developers are setting up by hand the parallel variables in the input file.

Let us know if it works properly.

Best,
D.

Ciao Daniele,

Thanks for the prompt reply. Another question that might seems stupid - but I was not able to find an answer. My log and report files do not include any information about the allocated or used memory. Is there a flag to have a more verbose output I should use in Yambo input file?

Thanks and all the best,
Roberto

Ciao Roberto,
in order to have memory information you should set a flag in compilation: Best,
Daniele

In the log file you should find messages when a variable which requires a significant amount of memory is allocated and de-allocated.

If you need more info you can compile yambo with the configure option
--enable-memory-profile (it just works in the last release I think)
You may even find useful to compile it with
--enable-time-profile

Best,
D.

Dear Davide and Daniele,

Thanks for your reply. I will try to see if I can see differences on how memory or workload are distributed. I will also compile YAMBO with that flag.

Best regards,
Roberto