Running Aview in Batch Mode On Headless Machine

I'm working with a heavily parameterized (lots of design variables) model and am trying to generate the adm/acf input files for a Monte Carlo through Adams Insight. Unfortunately the process of writing out the adm/acf input files for every Monte Carlo case (not even running the simulation) takes many hours (I suspect due to the number of design variables in the model). I was hoping to submit this process to our compute cluster to speed things up, but that system has no displays present.

With previous versions of Adams I was able to run aview in batch mode on a headless machine by connecting to a dummy Xvfb display (en.wikipedia.org/.../Xvfb), but Adams 2019.2 just throws these errors when I follow the same process (ip address replaced with XX.XXX.X.XX).

Tue Dec 15 12:46:02 2020 ERROR! pid[10285] failed to connect DS server.

Tue Dec 15 12:46:02 2020 ERROR! pid[10285] Failed to connect to DSServer, DS address="XX.XXX.X.XX".

Tue Dec 15 12:46:02 2020 SCA.SCASystemException - Failed to connect to DSServer, DS address="XX.XXX.X.XX". (ID=2028000000)

SCA.SCASystemException - Failed to connect to DSServer, DS address="XX.XXX.X.XX". (ID=2028000000)

Has anyone had any luck running recent version of aview headless? I have no issues running solver on the headless compute cluster but unfortunately the adm/acf generation process must be run inside aview.

Links (1)

https://en.wikipedia.org/wiki/Xvfb

0 Maziar R over 4 years ago

Hi Michael,

I am not quite familiar with this topic but I suppose reviewing the attached script can be helpful.

Thank you,

Maziar

Attached Files (1)

runXvfb.csh.txt
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 michael.w.clarke over 4 years ago

Hi Maziar,

Thanks for the quick reply. I tried the above script but its getting an error when it tries to run the rtool command with 2019.2:

"Data type/range violation. check Meta data definition"
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 Henrik.Skovbjerg over 4 years ago

You may have better luck using Adams 2020 FP1 (or the most recent 2021). Some fixes were made for running on non-GPU machines, and these were included in the Adams 2020 FP1 release.
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 Kent West over 4 years ago

Hi Michael,

Try setting this in your script before launching Adams View:

MSC_ADAMS_VIEW_SCA_REMOTING_OFF

Just set it to have a value of '1', perhaps. I hope that that takes care of the SCA errors.

Regarding the virtual framebuffer stuff & rtool: you can remove that rtool call in the script because I don't think that we have an X11 graphics driver anymore. I *think* that most virtual framebuffers now will work with the default 'OpenGL' setting, surprisingly..

I wonder if setting that env variable lets things work for you?

Fingers crossed,

Kent
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 Kent West over 4 years ago

quick update: Henrik tells me that you'll likely still have problems with 2019 and the latest releases are a better way to go here. Sorry if I've raised your hopes needlessly! Hope something works here for what you have available..
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 michael.w.clarke over 4 years ago

Thanks Henrik/Kent. That MSC_ADAMS_VIEW_SCA_REMOTING_OFF variable does eliminate the error messages but 2019.2 still doesn't launch. I'll give the 2020/2021 releases a shot once we get it downloaded.
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 Martin Kieltsch over 4 years ago

How about using Python code ?

I had done a benchmark exemple by setting 10000 parameter variables in a loop and doing the very same thing in Python gave me a speedup of about factor 350.

Code was like

for var = .acar.index start = 1 end = 10000

variable set variable_name=.acar.somevar real_value=10.99

end

or in python:

import aview_main

i=0

while i < 10000:

myCmd = "variable set variable_name=.acar.somevar real_value=%s" % 10.999

aview_main.execute_cmd(myCmd)

i+=1

If that works for you, it'd speed up your local file creation and you could get away without the batch issues on the cluster ?

Even if you're on the cluster you may start with mdi -c python xxx.py and avoid the batch mode. Maybe it'll work without those graphical issues ?
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 Ben Thornton over 1 year ago

Did anyone ever figure this out?

The following produces no aview.log file at all. It just shows "Begin execution of Adams View ..." and the returns to the selection menu.

```

#!/bin/bash

Xvfb :1 -pixdepths 8 16 24 -fp /usr/share/X11/fonts/misc/ -screen 0 1600x1200x24 -fbdir /usr/tmp &

export DISPLAY=127.0.0.1:1

export MSC_ADAMS_VIEW_SCA_REMOTING_OFF=1

xvfbPID=$!

/opt/hexagon/adams/2023_2/mdi -c aview ru-st f hello_world.cmd

kill $xvfbPID

unset DISPLAY

```
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel
0 Martin Kieltsch 6 months ago

For completeness's sake:
We had issues when starting ADAMS in background mode on a HPC cluster without a graphic card.
There was a bug in ADAMS that was fixed in version 2023.4 and higher.

I assume the above issue is affected, too, and could resolved by using ADAMS 2023.4 or higher.
Cancel
Vote Up 0 Vote Down

Sign in to reply

Verify Answer

Cancel