This is a known problem, here some workarounds:
Solution 1: Parallelize the BSE over the K-points e.g.:
Code: Select all
BS_CPU= "16 1 1" # [PARALLEL] CPUs for each role
BS_ROLEs= "k eh t"
Solution 2: Avoid to write database on file:
Solution 1, the more cpu you use, lesser files you will have in the SAVE dir.
Solution 2, preferred. Any file will be written on disk, so you have to diagonalize it on the same run. Drawbacks, any restart is possible.
In an internal devel version we have implemented a way to soften this problem by avoiding this huge fragmentation. We will insert it in the release very soon.