Issue with UDF in parallel processing

MC1 · August 28, 2017, 04:13

Hi there,

I have currently been running simulations which use a UDF to define inlet velocity and turbulenc kinetic energy profiles on my computer (using parallel processing w. 8 cores) without any issues.

After saving the case (having interpreted the UDF), I've been trying to run the simulations on a research computing cluster, though without success. The .jou file seems to bring in the case okay, but soon afterwards this happens:

> ;
; Autosave every 1000 iterations
file/auto-save/data-frequency 10000

> ;
; Run 10 iterations using first-order upwind
file/auto-save/root-name steady-firstorder

> solve/set/discretization-scheme/mom 0

Error: chip-exec: function "ABL_tke" not found.
Error: chip-exec: function "ABL_tke" not found.
Error: chip-exec: function "ABL_tke" not found.
Error: chip-exec: function "ABL_tke" not found.
Error: chip-exec: function "ABL_tke" not found.
Error: chip-exec: function "ABL_tke" not found.
Error: chip-exec: function "ABL_tke" not found.
Error: chip-exec: function "ABL_tke" not found.

Initialize using the hybrid initialization method.
...

I've read that this might be due to having to "parallelize" my UDF, though I'm not really sure how / why this works, and why when I run parallel processes on a single computer they seem to be okay, though when I submit a job to the cluster it does not work.

Can anyone pinpoint the cause of my error, and help me with a solution? I would be extremely grateful!

CeesH · August 28, 2017, 04:54

Are you sure this is related to parallel? What happens when you run the script in serial on your pc/cluster?

MC1 · August 28, 2017, 22:30

It doesn't seem to be working on serial either... not sure what might be the case though. My journal file currently reads:

define/user-defined/interpreted-functions "functionname.c" "cpp" 10000 no
;
file/read-case casename.cas
;
solve/iterate 10
file/write-case-data steady-firstorder
;
exit
yes

As I said, the UDF works fine when I run it on my PC, and the .c file is contained in the directory where I am running my case from.

CeesH · August 29, 2017, 02:53

So it seems to be an issue with loading it on your cluster. Do you use the same journal file when running it on your PC? Or do you then load the udf via the GUI?

MC1 · August 29, 2017, 03:01

I have just been loading the UDF using the GUI when running on my own PC. The cluster is a Linux-based HPC, and I've also ensured the line endings are UNIX style on the .c file to no effect.

CeesH · August 29, 2017, 03:17

maybe you can try loading the udf using the journal file on your workstation, the same way you do on your cluster. This would confirm that the error is indeed in the journal file. Unfortunately, it's been quite a while since I used journals to load UDFs (and if I did, it was compiled), so I can't quickly point out what the error might be.

August 28, 2017, 04:13	Issue with UDF in parallel processing	#1
MC1 New Member Mitchell Join Date: Aug 2017 Posts: 6 Rep Power: 8	Hi there, I have currently been running simulations which use a UDF to define inlet velocity and turbulenc kinetic energy profiles on my computer (using parallel processing w. 8 cores) without any issues. After saving the case (having interpreted the UDF), I've been trying to run the simulations on a research computing cluster, though without success. The .jou file seems to bring in the case okay, but soon afterwards this happens: > ; ; Autosave every 1000 iterations file/auto-save/data-frequency 10000 > ; ; Run 10 iterations using first-order upwind file/auto-save/root-name steady-firstorder > solve/set/discretization-scheme/mom 0 Error: chip-exec: function "ABL_tke" not found. Error: chip-exec: function "ABL_tke" not found. Error: chip-exec: function "ABL_tke" not found. Error: chip-exec: function "ABL_tke" not found. Error: chip-exec: function "ABL_tke" not found. Error: chip-exec: function "ABL_tke" not found. Error: chip-exec: function "ABL_tke" not found. Error: chip-exec: function "ABL_tke" not found. Initialize using the hybrid initialization method. ... I've read that this might be due to having to "parallelize" my UDF, though I'm not really sure how / why this works, and why when I run parallel processes on a single computer they seem to be okay, though when I submit a job to the cluster it does not work. Can anyone pinpoint the cause of my error, and help me with a solution? I would be extremely grateful!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Parallel processing problem	newbie29	OpenFOAM Running, Solving & CFD	1	June 22, 2012 04:23
OF Parallel Processing with Core i7 - How to Handle Hyperthreading	dancfd	OpenFOAM Running, Solving & CFD	5	March 5, 2012 21:43
bubbly flow and parallel processing	mvee	FLUENT	0	September 12, 2007 05:08
Help: how to realize UDF on parallel cluster?	Haoyin	FLUENT	1	August 6, 2007 13:53
UDF in Parallel	Gavin	FLUENT	1	June 19, 2005 22:17

August 28, 2017, 04:54		#2
CeesH Senior Member Cees Haringa Join Date: May 2013 Location: Delft Posts: 607 Rep Power: 0	Are you sure this is related to parallel? What happens when you run the script in serial on your pc/cluster?

August 28, 2017, 22:30		#3
MC1 New Member Mitchell Join Date: Aug 2017 Posts: 6 Rep Power: 8	It doesn't seem to be working on serial either... not sure what might be the case though. My journal file currently reads: define/user-defined/interpreted-functions "functionname.c" "cpp" 10000 no ; file/read-case casename.cas ; solve/iterate 10 file/write-case-data steady-firstorder ; exit yes As I said, the UDF works fine when I run it on my PC, and the .c file is contained in the directory where I am running my case from.

August 29, 2017, 02:53		#4
CeesH Senior Member Cees Haringa Join Date: May 2013 Location: Delft Posts: 607 Rep Power: 0	So it seems to be an issue with loading it on your cluster. Do you use the same journal file when running it on your PC? Or do you then load the udf via the GUI?

August 29, 2017, 03:01		#5
MC1 New Member Mitchell Join Date: Aug 2017 Posts: 6 Rep Power: 8	I have just been loading the UDF using the GUI when running on my own PC. The cluster is a Linux-based HPC, and I've also ensured the line endings are UNIX style on the .c file to no effect.

August 29, 2017, 03:17		#6
CeesH Senior Member Cees Haringa Join Date: May 2013 Location: Delft Posts: 607 Rep Power: 0	maybe you can try loading the udf using the journal file on your workstation, the same way you do on your cluster. This would confirm that the error is indeed in the journal file. Unfortunately, it's been quite a while since I used journals to load UDFs (and if I did, it was compiled), so I can't quickly point out what the error might be.