[Mesa-users] Switching MESA-r9798 to the newest version r24.3.01

Wed Jul 24 14:43:45 UTC 2024

hi rita,

mesa has always solved the coupled hydrodynamic, burn, and mixing equations as a single unit. it's one of mesa's unique strengths. see figure 47 and section B.2 of mesa II (paxton et al 2013) for a detailed description. that is, by default, there is no "network solver" in the sense of decoupling the burn from the hydro. 

i'll second eb's suggestion of posting a mesa work directory. its the only way anyone who might be interested in taking a look can potentially offer specific advice.

fxt

> On Jul 23, 2024, at 8:42 PM, Kit Yu Lau via Mesa-users <mesa-users at lists.mesastar.org> wrote:
> 
> Hi all,
> 
> We doubt that the running time is so slow because different usage of network solver (NOT network, but network solver for ns_h). Can anyone tell me what network solver is the newest version using? Thank you very much.
> 
> Regards,
> Rita
> 
> 
> On Wed, Jul 17, 2024 at 4:00 AM Ebraheem Farag <ekfarag at asu.edu <mailto:ekfarag at asu.edu>> wrote:
>> Hi Rita, 
>> 
>> Thanks for sharing your inlists!
>> 
>> Since you're modeling something quite specific and unique, I would advise against sharing disparate inlists where it is unclear when each is used, and where the loaded .mod file is not included. I would instead always recommend sharing a self contained clean model directory so it is easy for someone to unzip/untar ./clean, ./mk. and ./rn your model without worrying about how to put the pieces together. A clean directory should be <1Mb typically so it's relatively easy to share. shmesa <https://github.com/MESAHub/mesa/tree/main/scripts/shmesa> is now included with MESA r24.03.1+ which contains commands to do this for you, although it can be done by hand relatively easily as well. 
>> 
>> Nonetheless, since you're using rp.net <http://rp.net/> with 303 isotopes, I can imagine this is quite an expensive run. I think the biggest clues will lie in whether your model is struggling to find solutions. Can you share the terminal output (as a singular txt file) from your runs in both versions, as I do not have the bandwidth to test these inlists. We can directly compare the terminal outputs to see if anything particularly different is happening for both runs, which should give some insight into the differences. 
>> 
>> Another extremely useful way to diagnose any issues or differences would be to generate pgstar pngs every 1-10 timesteps and then compress them into a movie (e.g. "images_to_movie "png/*.png" movie.mp4"). The pgstar movie is an extremely useful diagnostic for numerics, not just physics! See https://docs.mesastar.org/en/release-r24.03.1/test_suite/ns_h.html. In fact I often use pgstar to diagnose issues for runs that have to be done on a cluster, so there is a visible track record of what my model has done that I can review after the fact, and before post-processing or analyzing my results for their scientific merit. as far as your plots go, they look good, but I have questions like "how many model numbers did this take, how many retries, how many solver iterations", things that are not communicated by this plot but relate more directly to your question of "why is my model running slow?". It could be useful also to compare your models and their run time to the test_cases ns_h, ns_he, as you can switch the test cases to the larger network you're adopting and see if this issue is related to your inlist setup in the current MESA version or not.
>> 
>> Lastly, OMP_NUM_THEADS=64 sounds like a lot of cores, which might not be giving you too much of a speed up, and could potentially slow your model down. MESA does not parallelize very well past 12-16 threads, so it might be worth saving some cluster resources and reducing the number of threads? I personally used 16 for most models, so it would be interesting to know if your model runs faster or slower with OMP_NUM_THREADS=16 as opposed to 64?
>> 
>> I hope this helps!
>> 
>> -EbF
>> 
>> On Tue, Jul 16, 2024 at 7:27 AM Kit Yu Lau via Mesa-users <mesa-users at lists.mesastar.org <mailto:mesa-users at lists.mesastar.org>> wrote:
>>> Hi all,
>>> 
>>> Thanks for replying to my emails.  I attached the inlist files and run_star_extras.f90 files in the email. My OMP_NUM_THREADS is 64. Thank you!
>>> 
>>> Regards,
>>> Rita
>>> 
>>> On Tue, Jul 16, 2024 at 8:54 PM Francis Timmes <fxt44 at mac.com <mailto:fxt44 at mac.com>> wrote:
>>>> hi rita,
>>>> 
>>>> thanks for the plot - looks reasonable. i will repeat eb's suggestion to post your inlist(s) and any associated run_star_extras, otherwise its impossible to offer any suggestions for potentially improving the run time. in addition, please give the value of your OMP_NUM_THREADS environment variable.
>>>> 
>>>> fxt
>>>> 
>>>> 
>>>> 
>>>> 
>>>> 
>>>>> On Jul 16, 2024, at 1:08 AM, Kit Yu Lau via Mesa-users <mesa-users at lists.mesastar.org <mailto:mesa-users at lists.mesastar.org>> wrote:
>>>>> 
>>>>> This is the luminosity plot I got for the new version. It takes 1  week to get the following resutls:<image.png>
>>>>> 
>>>>> On Tue, Jul 16, 2024 at 12:29 AM Ebraheem Farag <ekfarag at asu.edu <mailto:ekfarag at asu.edu>> wrote:
>>>>>> Hello Kit,
>>>>>> 
>>>>>> Please share your inlists or some sort of output comparison (perhaps attach txt files of the terminal output from both of your runs or the logs or a pgstar) if you want comment. It's nearly impossible to speculate otherwise except to say that the current version of MESA is much more careful  in a variety of respects and takes extra solver iterations to ensure good numerical convergence with respect to previous versions including the use energy conservation to maintain numerical accuracy (something not done in mesa-r9798). Much of this is documented in MESA IV <https://urldefense.com/v3/__https://ui.adsabs.harvard.edu/abs/2018ApJS..234...34P/abstract__;!!IKRxdwAv5BmarQ!a123359XKv1btj2551_X1jOESvLwTfWENBGaMxuC1YXMIsVGlvfTQGkxleBAQeXzmPiaVX75ND4fs8ReScK_KXftMcH-QA$> and MESA V <https://urldefense.com/v3/__https://ui.adsabs.harvard.edu/abs/2019ApJS..243...10P/abstract__;!!IKRxdwAv5BmarQ!a123359XKv1btj2551_X1jOESvLwTfWENBGaMxuC1YXMIsVGlvfTQGkxleBAQeXzmPiaVX75ND4fs8ReScK_KXc7sFpJyQ$>.
>>>>>> 
>>>>>> -EbF
>>>>>> 
>>>>>> On Mon, Jul 15, 2024 at 6:40 AM Kit Yu Lau via Mesa-users <mesa-users at lists.mesastar.org <mailto:mesa-users at lists.mesastar.org>> wrote:
>>>>>>> Dear all, 
>>>>>>> 
>>>>>>> I used the most updated version of mesa to run X-ray bursts. Comparing with the old version (mesa-r9798) . It is much slower. I spent 10 days for the new version to get 2 bursts but the old version I spent 3 days to get 5 bursts. All the input files are the same for both versions. Is there any way to speed up the new version? Thank you very much.
>>>>>>> 
>>>>>>> Regards,
>>>>>>> Rita Lau
>>>>>>> 
>>>>>>> On Wed, Jun 19, 2024 at 1:48 PM Frank Timmes <fxtimmes at gmail.com <mailto:fxtimmes at gmail.com>> wrote:
>>>>>>>> hi rita,
>>>>>>>> 
>>>>>>>> mesa has always used an evolving patchwork of eos modules to cover the rho-t plane. this eos patchwork for recent versions is shown in figure 7 of section 4 in mesa vi (jermyn et al 2023), also described at https://docs.mesastar.org/en/release-r24.06.1-rc1/eos/overview.html <https://urldefense.com/v3/__https://docs.mesastar.org/en/release-r24.06.1-rc1/eos/overview.html__;!!IKRxdwAv5BmarQ!YEDGoBxIue5Bt2R-BRTkPSG-6fl2vcTcqMub_4ovLO-Q0TQt9hld6JxESGO8MaIX13vdzJfkkDBMN5Tyzs2O_6pMof9Osw$> with inlist controls at https://docs.mesastar.org/en/release-r24.06.1-rc1/eos/defaults.html#eos-module-controls <https://urldefense.com/v3/__https://docs.mesastar.org/en/release-r24.06.1-rc1/eos/defaults.html*eos-module-controls__;Iw!!IKRxdwAv5BmarQ!YEDGoBxIue5Bt2R-BRTkPSG-6fl2vcTcqMub_4ovLO-Q0TQt9hld6JxESGO8MaIX13vdzJfkkDBMN5Tyzs2O_6qY4lMFmA$> . the current eos patchwork cannot be easily changed to the eos patchwork used in 2017 for r9798.
>>>>>>>> 
>>>>>>>> fxt
>>>>>>>> 
>>>>>>>> 
>>>>>>>> 
>>>>>>>> 
>>>>>>>> 
>>>>>>>>> On Jun 17, 2024, at 11:23 PM, Kit Yu Lau via Mesa-users <mesa-users at lists.mesastar.org <mailto:mesa-users at lists.mesastar.org>> wrote:
>>>>>>>>> 
>>>>>>>>> Hi everyone,
>>>>>>>>> 
>>>>>>>>> I was using mesa-r9798 and I am switching to the newest version. However, there was no eos in mesa-r9798. I think I was using helm_table.dat for the old version. How do I change the newest version to use the same eos? I tried not to command anything in eos and it seems that it is using eos_free. How do I change the eos to helm_table.dat? Thank you very much.
>>>>>>>>> 
>>>>>>>>> Regards,
>>>>>>>>> Rita
>>>>>>>>> _______________________________________________
>>>>>>>>> mesa-users at lists.mesastar.org <mailto:mesa-users at lists.mesastar.org>
>>>>>>>>> https://lists.mesastar.org/mailman/listinfo/mesa-users <https://urldefense.com/v3/__https://lists.mesastar.org/mailman/listinfo/mesa-users__;!!IKRxdwAv5BmarQ!YEDGoBxIue5Bt2R-BRTkPSG-6fl2vcTcqMub_4ovLO-Q0TQt9hld6JxESGO8MaIX13vdzJfkkDBMN5Tyzs2O_6oc-vzPzw$>
>>>>>>>>> 
>>>>>>>> 
>>>>>>> _______________________________________________
>>>>>>> mesa-users at lists.mesastar.org <mailto:mesa-users at lists.mesastar.org>
>>>>>>> https://urldefense.com/v3/__https://lists.mesastar.org/mailman/listinfo/mesa-users__;!!IKRxdwAv5BmarQ!YEDGoBxIue5Bt2R-BRTkPSG-6fl2vcTcqMub_4ovLO-Q0TQt9hld6JxESGO8MaIX13vdzJfkkDBMN5Tyzs2O_6oc-vzPzw$ 
>>>>>>> 
>>>>> _______________________________________________
>>>>> mesa-users at lists.mesastar.org <mailto:mesa-users at lists.mesastar.org>
>>>>> https://lists.mesastar.org/mailman/listinfo/mesa-users <https://urldefense.com/v3/__https://lists.mesastar.org/mailman/listinfo/mesa-users__;!!IKRxdwAv5BmarQ!a123359XKv1btj2551_X1jOESvLwTfWENBGaMxuC1YXMIsVGlvfTQGkxleBAQeXzmPiaVX75ND4fs8ReScK_KXf0l7ozvg$>
>>>>> 
>>>> 
>>> _______________________________________________
>>> mesa-users at lists.mesastar.org <mailto:mesa-users at lists.mesastar.org>
>>> https://urldefense.com/v3/__https://lists.mesastar.org/mailman/listinfo/mesa-users__;!!IKRxdwAv5BmarQ!a123359XKv1btj2551_X1jOESvLwTfWENBGaMxuC1YXMIsVGlvfTQGkxleBAQeXzmPiaVX75ND4fs8ReScK_KXf0l7ozvg$ 
>>> 
> _______________________________________________
> mesa-users at lists.mesastar.org
> https://lists.mesastar.org/mailman/listinfo/mesa-users
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20240724/52c89e69/attachment.htm>