My Community

Hello. I've been running quite a long MD of a solvated ion pair in a cubic box and it seems that the cpu time required for an iteration in the electronic relaxation ("RMM") (ALGO=fast, LREAL=A) increases. The increase occurs in two abrupt jumps, one at ~2000 MD steps and another at ~5500 steps, resulting in almost twice as long cpu times (grep LOOP OUTCAR) as at the beginning of the run. After that the cpu times remain more or less the same, even after tens of thousands of steps.

Interestingly, virtually the same cpu time pattern occurs with different simulations of the same system, e.g. when I restart previously terminated run or when I change ALGO to 38, etc -- the LOOP timings are short at the beginning, but jump up in two bursts, then remain stable.

Is there any reasonable explanation for this and/or a way to avoid increased cpu times? -- Thanks!

Hi,

are you sure you are alone on the machine?

Cheers,

alex

alex wrote: are you sure you are alone on the machine?

Yes, I'm pretty sure about that. No trace of other users or worthwhile system tasks. Also, the fact that the pattern of increasing cpu times is virtually the same in several independent simulations prevents me from believing the increase is due to other users taking part of cpu resources.

This is very weird, indeed. Some things to check: have you tried

- locating a faulty network adapter or
- does your filesystem cache run out of space (with lenghty outputs, but I doubt it'd be a factor of two)
- have you checked if the additional time is user/system or whatever (e.g. at the end of OUTCAR)
- have you run out of physical memory (e.g. are you using swap)
- ...

Good luck! And keep us posted, please!

Cheers,

alex