New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
runtime.dat no longer signals the job to stop #1304
Comments
Hi, I have to admit I did not use that feature yet but In case you have a minimal example please attach it and the name the code version it works with. You can use git bisect to pinpoint the commit yourself, that should be the quickest and easiest solution. best tobi |
I think we lost this functionality around version 7 (at the time I did not know we had it) maybe because it was complicated to guarantee the right file outputs when the simulation is stopped this way, especially for discrete adjoint problems. |
I haven't looked at the python wrapper in detail, we had previously tracked several properties which are not available in the conventional convergence tracker (such as Cmy) and used the stop file to obtain the data when appropriate. We have considered increasing the save frequency, but there are several files which are not created at every save point; thereby forcing each "standard" run to become a "run to nearest save" then "resume for 1 time-step to get the actual output data". As the decomposition and some of the file writes take a significant amount of time, it is a non-starter for commercial applications. We would switch back to v6 (which had the capability) but we need some of the other features in v7. I'll take a look at the python wrapper. Thank you for your candor. |
Can you describe a bit more the stopping strategy you want? The new output structure is much more flexible than before, and you can even use Cauchy criteria based on time averages (of pretty much any coefficient), with different windowing methods and so on. |
I know updating versions can be a pain (even for us developers) but it is truly worth it in this case. |
Being able to apply Cauchy criteria to any of the named parameters would be wonderful. I see that some of the options are present in the config_template; is there a list of name conversions somewhere? (i.e. "DRAG" -> CD, "Cmy" -> ?, and I assume TAVG_ applies an average over CONV_WINDOW_CAUCHY_ELEMS elements for the requested parameter; I see some at https://su2code.github.io/docs_v7/Custom-Output/, but it is listed as non-exhaustive) Beyond that, there are some instances we run into where the solution parameter does not converge according to the Cauchy criterion but it is converged for our uses (pitching moment convergence can be a bit different from other forms of convergence; it can sometimes wander within a count). In those instances, we would like to be able to terminate the job without incurring the "wait until save and run it again" penalty. |
Does Cauchy criteria based on time averages only work for time-independent flows? Because it seems that WINDOW_START_ITER needs to be set in the section of TIME-DEPENDENT SIMULATION in the input .cfg file. Thank you. |
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If this is still a relevant issue please comment on it to restart the discussion. Thank you for your contributions. |
For those who find this via google - resolution options: 1) see if any anyone has restored the functionality, or 2) use the python wrapper |
Customization should be done via python and communicating via files is a bit of hack anyway, requires extra MPI barriers, etc. etc. ugly. |
Hello,
In version 7.1.0, runtime.dat no longer functions to stop the job using "OUTER_ITER=", "EXT_ITER=", or "TIME=". Did the parameter change?
(CentOS 7, gluster FS, GCC 7.3.1)
The text was updated successfully, but these errors were encountered: