SSCHA
uct_md.c:203 UCX ERROR #338
-
|
Dear SSCHA developers, I'm a new SSCHA user. I'm doing a Au example on SSCHA website "[Tutorial 01 - First Simulations (https://sscha.eu/Tutorials/tutorial_01_first_simulations/)". [1717677221.254829] [n27:162451:0] uct_md.c:203 UCX ERROR Transport 'lx5' does not exist I wonder why these erros appear. Best regards, |
Beta Was this translation helpful? Give feedback.
Replies: 4 comments 29 replies
-
|
Hi Lin, This is the first time I have seen these errors, and we need more details to understand what went wrong. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for your reply. |
Beta Was this translation helpful? Give feedback.
-
|
I find that there is a remind after installing with Anaconda: On Linux, Open MPI is built with UCX support but it is disabled by default. On Linux, Open MPI is built with CUDA awareness but it is disabled by default. I dont whether erros come from this problem. |
Beta Was this translation helpful? Give feedback.
-
|
Mr.
I find that I perform the commond "pip install cellconstructor python-sscha tdscha" when I install SSCHA. |
Beta Was this translation helpful? Give feedback.
-
|
Are you submitting the job in parallel? The error here is from matplotlib, as unveiled from the last lines of your error: You can manually remove the file, it says. Indeed, if you run the job in parallel, it is possible that multiple process created that file, thus leading to the error. This should not occur, but it is most likely due to an error of how MPI is configured. Try removing that file (if any) and rerun it with 1 processor. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, I run in parallel. However, there is not a file named "fontlist-v390.json.matplotlib-lock" but a file named "fontlist-v390.json" in the folder ".cache/matplotlib/". |
Beta Was this translation helpful? Give feedback.
-
|
Ok, let me know if on a single processor it works. |
Beta Was this translation helpful? Give feedback.
-
|
It's really that erros result from not correctly setting parallelization as you say. |
Beta Was this translation helpful? Give feedback.
-
|
@mesonepigreco |
Beta Was this translation helpful? Give feedback.
-
|
I see. In a separate iteration, SSCHA minimization is done after QE calculations, and then a second iteration is performed. Supposing that I use SLURM system to apply for 50 processors, and the script does SSCHA minimization with these 50. Between SSCHA minimization steps, QE calculations use 8 of these 50. |
Beta Was this translation helpful? Give feedback.
-
|
That's the case only if you do not use the cluster function for calling the QE from SSCHA. If you make a manually QE call from SSCHA then you will use the processors declared in the SLURM run. On the other hand if you use the cluster functionality, then the QE calculations are processed as a serie of different SLURM calls and are independent from the main SSCHA script. Check the http://sscha.eu/Tutorials/tutorial_02_advanced_submission/ |
Beta Was this translation helpful? Give feedback.
-
|
That's it! |
Beta Was this translation helpful? Give feedback.
-
|
@biglinn The new version (1.5), which will be released very soon, introduces a new mode that enables running the SSCHA on the cluster with a long serial job. Then, that job proceeds by submitting other jobs with the espresso calculations directly to the cluster. This second mode is very useful if the SSH connection to the cluster is difficult to maintain (e.g., we do not have a workstation, or the connection to the cluster must occur via a 2FA protocol that requires human interaction). |
Beta Was this translation helpful? Give feedback.
-
|
Oh, that may be my problem! I exactly submit the SSCHA script on a cluster. So I'll study the available mode. |
Beta Was this translation helpful? Give feedback.
-
|
@mesonepigreco and @diegomartinez2 This script "ensenmble.py" is as follows: import cellconstructor as CC, cellconstructor.Phonons
import sscha, sscha.Ensemble
import numpy as np
# Fix the seed so that we all generate the same ensemble
np.random.seed(0)
# Load the dynamical matrix
dyn = CC.Phonons.Phonons("../ktao3_harmonic_dyn", nqirr=19)
dyn.Symmetrize()
#[ apply here the needed changes to dyn ]
# Prepare the ensemble
temperature = 300 # 300 K
ensemble = sscha.Ensemble.Ensemble(dyn, temperature)
# Generate the ensemble
number_of_configurations = 10
ensemble.generate(number_of_configurations)
# Save the ensemble into a directory
save_directory = "data"
population_id = 1
ensemble.save(save_directory, population_id) |
Beta Was this translation helpful? Give feedback.
-
|
Yes, you should also check if the gradients reach zero within the stochastic noise. |
Beta Was this translation helpful? Give feedback.
-
|
Hi, @mesonepigreco However, the program reports a erro:
|
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
There is a mismatch between the size of the data loaded in the ensemble (the supercell size) and the one of the dynamical matrix used to initialize the ensemble. The line you mention initializes the ensemble with the provided dynamical matrix and at the given temperature (0 K). This also sets the number of atoms expected when you call the load function and could be the reason behind the mismatch. Try to use the same dynamical matrix employed to generate the ensemble. Anyway, consider opening a new discussion for follow-up questions, as they are no longer related to the title and could get confusing for other people. |
Beta Was this translation helpful? Give feedback.
-
|
Okay, thanks! I'll open a new discussion if I have another problem. |
Beta Was this translation helpful? Give feedback.
Are you submitting the job in parallel?
The error here is from matplotlib, as unveiled from the last lines of your error:
You can manually remove the file, it says. Indeed, if you run the job in parallel, it is possible that multiple process created that file, thus leading to the error. This should not occur, but it is most likely due to an error of how MPI is configured.
Try removing that file…