The OPAL Discussion Forum

[Christof Metzger-Kraus <christof.j.kraus AT gmail.com>]

Dear all,

the increased compute time that both Chris and Nicole observed originate from two different sources: Chris used a traveling wave structure with the parameter FAST=FALSE. In Opal version 2.2 this parameter had no effect but since version 2.4 it does. After switching to FAST=TRUE he gets comparable compute times for both versions. The problem with Nicoles simulation is that three particles travel backwards away from the bunch. Opal-T unlike Opal-cycl doesn't delete these particles (yet). Together with a space charge solver that splits the discretization grid in longitudinal direction only this results in a very imbalance in load between the cores. After adding the necessary code to delete such particles I got a well balanced load again. I'll soon add this changes to the source code repository.

Best,

Christof

On Sun, Jan 17, 2021 at 7:09 AM Nicole R Neveu <nneveu AT stanford.edu> wrote:

Hi guys,

 

I needed to do a rebuild today on a computer I wasn’t using for a few months.

I saw a similar behavior to Chris + one more difference.

When I re-ran a LCLS-II test with 2.4, the run time was 9 minutes vs. 6 minutes on my old 2.0 build.

 

I went back and built off the 2.2 branch, and the file would not run.

The error was related interpolation when loading the last field map (a standing wave rf cavity).

See file attached.

 

Then I went back to a simpler test (AWA gun on regression tests).

Only one rf cavity in that file (the gun) + solenoids, and there was no time difference or error when switching between 2.2 and 2.4?

Could this point to something field map related?

 

Thanks,

 

Nicole

 

From: <opal-request AT lists.psi.ch> on behalf of Christof Metzger-Kraus <christof.j.kraus AT gmail.com>
Reply-To: Christof Metzger-Kraus <christof.j.kraus AT gmail.com>
Date: Saturday, January 16, 2021 at 7:18 AM
To: Chris Hall <chall AT radiasoft.net>
Cc: opal <opal AT lists.psi.ch>
Subject: Re: [Opal] Increase in External Field Evaluation Time

 

Hi Chris,

 

I've ran our regression tests with version 2.2 and 2.4 of Opal but I didn't notice any dramatic increase in compute time for the evaluation of external fields for any test. Some ran a bit slower others a bit faster. Can you share the input file and the fields maps? If not: what kind of field maps do you use? Do you use other features such as wake fields or particle matter interaction?

 

Christof

 

On Tue, Jan 12, 2021 at 6:21 AM Christof Metzger-Kraus <christof.j.kraus AT gmail.com> wrote:

Hi Chris,

 

I can't find anything in the changes of the source code that could have such an influence on the compute time for the external fields. I'll investigate this.

 

Christof

 

On Mon, Jan 11, 2021 at 9:44 PM Chris Hall <chall AT radiasoft.net> wrote:

Hi All,

 

I recently noticed that some OPAL simulations were taking much longer than expected. We upgraded to OPAL 2.4.0 from 2.2.0 not that long ago so I reran an old simulation I still had outputs for. From this I see that the average Wall and CPU times for "External field eval." have increased by almost a factor of 10 when I compare 2.4 to 2.2. This leads to a doubling of the total wall clock run time. None of the other individual timing results show significant difference between the two runs.

 

Is anyone aware of changes between the two versions that might have caused this, or any other factors that should be investigated that would lead to this?

 

I've attached the run logs (with step updates cut out) for the runs on versions 2.2 and 2.4 for reference.

 

Thanks!

--

Chris hall

Research Scientist | RadiaSoft

720-502-3928 x709 | chall AT radiasoft.net

radiasoft.net | sirepo.com