Resource reqirement

[DSarkar]

Hi,

Is there a way to estimate the time required to finish a simulation given the computation resources for a certain simulation setup? It would be helpful for me to estimate the time required as I am running on a common queued cluster, so that I can request resources optimally.

Regards,

Debarghya

[Ulrik G. Vej-Hansen]

Hi Debarghya

Unfortunately, we do not have a way of automatically estimating the time required for a calculation. However, if you have calculations for similar systems you can use those to estimate the time required. If you only change a few things, it will probably not change the time much. However, you should always be skeptical when doing such estimates - the other calculation might take a lot longer than expected.

If your calculation involves geometry optimization, you can use that the time required for each force evaluation step is approximately the same for a given system.

[DSarkar]

Hi Ulrik,

Thanks for letting me know.
With respect to your comment: "If your calculation involves geometry optimization, you can use that the time required for each force evaluation step is approximately the same for a given system." How can we know the time taken for each step? Could you please elaborate a little more?

Regards,

Debarghya

[Ulrik G. Vej-Hansen]

Well, that is the challenge in this case, as it is not possible to know beforehand. If you have a Geometry Optimization which has been interrupted, you can restart from the trajectory and look at the time per force step when estimating how much longer the simulation will run.

The absolute best help when estimating the length of a simulation is experience, and you should always allow for a lot more time than estimated. Personally I would set the time limit of at least twice my initial estimate unless I was very familiar with the system I was studying. For new systems I would give it at least 3 times longer than estimated.

Of course, all of this assumes that your supercomputer uses only actual running time for accounting. In this case I would say that it is always better to have too high estimate, to increase the chance of your calculation actually finishing.