FDS simulation run time estimation

1,281 views
Skip to first unread message

Emanuele Gissi

unread,
Oct 20, 2016, 7:20:28 AM10/20/16
to FDS and Smokeview Discussions
How do you estimate the simulation run time?
Using Wall clock and Total time from the *.out file?

I really miss the Runtime Estimator from Kris Overholt and I want to implement it back in BlenderFDS (http://firetools.github.io/blenderfds/).
Open source, of course.

Thanks in advance.
Emanuele

Randy McDermott

unread,
Oct 20, 2016, 7:25:18 AM10/20/16
to FDS and Smokeview Discussions
Hi Emanuele,

Try using the _cpu.csv file.

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+unsubscribe@googlegroups.com.
To post to this group, send email to fds...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/0f5443b9-08d9-41aa-9bfc-18deacc7189b%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Emanuele Gissi

unread,
Oct 21, 2016, 8:17:08 AM10/21/16
to FDS and Smokeview Discussions
Hi Randy,
thank you for the indication.

In fact I wish I could estimate the completion wall clock time of the simulation.
This is becoming important to forsee the cost of the calculation when using external services as Amazon EC2.

The following is the strategy I wish to employ.

1) Estimate the maximum fluid velocity (from HRR), then use the CFL constraint and mesh cell size to calculate the time step size.

2) By using the wall clock from the CHID.out file, estimate early during the simulation the wall clock time per time step (very dependent on hardware and case complexity), that seems to stay rather constant.

3) Calculate:
(T_END - T_BEGIN) / time step * wall clock per time step = estimated calc duration
The result should overshot a little because at start the fluid velocity is less than max. 

What do you think? Does it make sense?

Best regards,
Emanuele

Il giorno giovedì 20 ottobre 2016 13:25:18 UTC+2, Randy McDermott ha scritto:
Hi Emanuele,

Try using the _cpu.csv file.
On Thu, Oct 20, 2016 at 7:20 AM, Emanuele Gissi <emanuel...@gmail.com> wrote:
How do you estimate the simulation run time?
Using Wall clock and Total time from the *.out file?

I really miss the Runtime Estimator from Kris Overholt and I want to implement it back in BlenderFDS (http://firetools.github.io/blenderfds/).
Open source, of course.

Thanks in advance.
Emanuele

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.

Randy McDermott

unread,
Oct 21, 2016, 8:26:15 AM10/21/16
to FDS and Smokeview Discussions
Hi Emanuele,

That sounds like a good plan.  My only comment/question would be, if are making use of CHID.out anyway, then you have the time step, no need to estimate it, right?

To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+unsubscribe@googlegroups.com.

To post to this group, send email to fds...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/ed8e7b61-6f7f-4cad-8603-776e9fdf4d19%40googlegroups.com.

Emanuele Gissi

unread,
Oct 21, 2016, 8:41:24 AM10/21/16
to fds...@googlegroups.com
Hi Randy,
the problem is that early during the simulation the time step size has still not settled to its minimum (speeds still not fully developed). Fully development can happen after several days of calculation.

If I used those early values of the time step size I would badly under shot the total duration.

This is why I am proposing to estimate the minimum time step that will impact the most the real duration.

Sounds likely?

Emanuele

--
You received this message because you are subscribed to a topic in the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/fds-smv/u7TBfuA6LXs/unsubscribe.
To unsubscribe from this group and all its topics, send an email to fds-smv+unsubscribe@googlegroups.com.

To post to this group, send email to fds...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/CAC6wax1nBBT-0AaeEmSj4hFrfYpv1dgRP0o8X-mU-noE4DZfPg%40mail.gmail.com.

For more options, visit https://groups.google.com/d/optout.



--
http://www.emanuelegissi.eu/

Randy McDermott

unread,
Oct 21, 2016, 8:52:08 AM10/21/16
to FDS and Smokeview Discussions
Sure.  So, you will take the HRR from the input file and apply some sort of plume correlation for velocity?

To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/CAL0r%2BVnbxv4m4-gvti%3D%2BoG-EM4k5w%3DiW3RyCYfj6RP1ZhpjydQ%40mail.gmail.com.

Emanuele Gissi

unread,
Oct 21, 2016, 8:59:59 AM10/21/16
to fds...@googlegroups.com
Yes.

Then:
time step size = 0.80 · min cell size / estimated plume velocity

Right?
Emanuele


To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/CAC6wax1bYjhMF%3DtsU0i8uwdyAJ4c7jfByHcW3t6r7VsYgUtxfA%40mail.gmail.com.

For more options, visit https://groups.google.com/d/optout.



--
http://www.emanuelegissi.eu/

Randy McDermott

unread,
Oct 21, 2016, 9:14:04 AM10/21/16
to FDS and Smokeview Discussions
Sounds good.

To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/CAL0r%2BVmcTvnTeQ2m%3DYPRehQm9H5cnKJb2%3D%3DvXApSTY_iDZHY%3DQ%40mail.gmail.com.

dr_jfloyd

unread,
Oct 21, 2016, 9:22:14 AM10/21/16
to FDS and Smokeview Discussions
I think the approach will provide a reasonable ballpark estimate on run time for many basic types of simulations. There are a some cases where it might underestimate the time:

The CFL might be limited by flow through VENTs, flows induced by sprinklers, or other phenomena that a plume correlation would not capture.
Sprinkler operation later in a run could result in an increase in the CPU time per time step.  
For simple, well ventilated pool fires the cost of the combustion model will be low (only being done in a small number of cells) and it won't change much over the simulation. For simulations where the fire becomes vitiated and combustion is occurring in a large number of grid cells, the cost of the combustion model will increase over time as the number of cells with combustion increases. This won't be captured with an early query of the .out file.

--
You received this message because you are subscribed to a topic in the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/fds-smv/u7TBfuA6LXs/unsubscribe.
To unsubscribe from this group and all its topics, send an email to fds-smv+u...@googlegroups.com.

To post to this group, send email to fds...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.
To post to this group, send email to fds...@googlegroups.com.

--
You received this message because you are subscribed to a topic in the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/fds-smv/u7TBfuA6LXs/unsubscribe.
To unsubscribe from this group and all its topics, send an email to fds-smv+u...@googlegroups.com.

To post to this group, send email to fds...@googlegroups.com.

Emanuele Gissi

unread,
Oct 21, 2016, 10:11:39 AM10/21/16
to fds...@googlegroups.com

The real time step size from CHID.out could be used at a later stage of the simulation to improve the estimate.

Do you suggest any other possibility for improving the estimate?

Emanuele


To unsubscribe from this group and all its topics, send an email to fds-smv+unsubscribe@googlegroups.com.

To post to this group, send email to fds...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/869eceff-ea5e-4166-a8f6-40367553c61b%40googlegroups.com.

Dave McGill

unread,
Oct 21, 2016, 12:09:56 PM10/21/16
to FDS and Smokeview Discussions
Hi Jason,

I have a question regarding this part of your post.

On Friday, 21 October 2016 09:22:14 UTC-4, dr_jfloyd wrote:
....
For simple, well ventilated pool fires the cost of the combustion model will be low (only being done in a small number of cells) and it won't change much over the simulation. For simulations where the fire becomes vitiated and combustion is occurring in a large number of grid cells, the cost of the combustion model will increase over time as the number of cells with combustion increases. This won't be captured with an early query of the .out file.

In this post, Randy said, "The code goes through the same amount of work, more or less, no matter what the "activity" is in the simulation."  

Does the combustion activity have an affect upon simulation speed?

Thanks

Dave

Randy McDermott

unread,
Oct 21, 2016, 12:18:21 PM10/21/16
to FDS and Smokeview Discussions
The amount of work per time step is roughly the same.  But the time step can go down significantly due to combustion.

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+unsubscribe@googlegroups.com.

To post to this group, send email to fds...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/273850c4-731d-4532-8be2-ad7ffa21a636%40googlegroups.com.

Dave McGill

unread,
Oct 21, 2016, 12:24:24 PM10/21/16
to FDS and Smokeview Discussions
Randy,

So, the calculation in a given cell is essentially the same, regardless of whether or not combustion is present, but if combustion is involved the time steps will shorten which will slow down the simulation?

Thanks

Dave


On Friday, 21 October 2016 12:18:21 UTC-4, Randy McDermott wrote:
The amount of work per time step is roughly the same.  But the time step can go down significantly due to combustion.
On Fri, Oct 21, 2016 at 12:09 PM, Dave McGill <dwt.m...@gmail.com> wrote:
Hi Jason,

I have a question regarding this part of your post.

On Friday, 21 October 2016 09:22:14 UTC-4, dr_jfloyd wrote:
....
For simple, well ventilated pool fires the cost of the combustion model will be low (only being done in a small number of cells) and it won't change much over the simulation. For simulations where the fire becomes vitiated and combustion is occurring in a large number of grid cells, the cost of the combustion model will increase over time as the number of cells with combustion increases. This won't be captured with an early query of the .out file.

In this post, Randy said, "The code goes through the same amount of work, more or less, no matter what the "activity" is in the simulation."  

Does the combustion activity have an affect upon simulation speed?

Thanks

Dave

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.

To post to this group, send email to fds...@googlegroups.com.

dr_jfloyd

unread,
Oct 21, 2016, 12:45:44 PM10/21/16
to FDS and Smokeview Discussions
The combustion routine first checks to see if there is unburned fuel present in a cell. If there isn't it skips the combustion model for that cell.  All else being equal, a well ventilated fire that stays compact (e.g. few cells with fuel) will take somewhat less time per time step than a poorly ventilated one with fuel everywhere in the domain. 

Dave McGill

unread,
Oct 21, 2016, 12:49:39 PM10/21/16
to FDS and Smokeview Discussions
Ah, that's something I didn't know. 

Thanks
Reply all
Reply to author
Forward
0 new messages