Jump to content
Sign in to follow this  
crow9350

setting up Flip distributed Simulation on a single machine

Recommended Posts

How can I set up Flip distributed Simulation on a Single machine?

and do I need Houdini HQ?

Houdini Master Class demonstrates distribution sim on multiple machines, so I have no idea how to set it up in a single machine.

I know it requires more memory, but what I'm trying to do is too big so it seems impossible to do it with just regular setting.

I anticipate 1gb of file size of fluid or white water per frame if I do it with regular setting, 

I think it would take up to 2 to 3weeks on my computer. and I'm not sure my computer can handle it.

(my cpu is amd ryzen2700, ram64gb, windows10)

Do you think it is still better without distribution?

Share this post


Link to post
Share on other sites

I think you need to do more simple project or find the way to simplify current. Distribution on one machine have no profits.

Share this post


Link to post
Share on other sites

Yeah, " find the way to simplify current" was the last thing wanted to do. but, it seems that I have to change my mind. thanks. 

Edited by crow9350

Share this post


Link to post
Share on other sites
On 25/06/2020 at 7:07 PM, tamagochy said:

I think you need to do more simple project or find the way to simplify current. Distribution on one machine have no profits.

What if you need to do a flip sim that requires 10 km and 100 million particles but can only sim 20 million before your machine crashes?  Can you use distributed with/without HQueue to slice it up and just sim 5x on the one machine, but in the end have a nice 100 mill sim? 

Edited by HM_2020

Share this post


Link to post
Share on other sites

Distribution on a single machine is only really useful for faster sim times. There are a couple of parts of the FLIP solver which are still single threaded.
 

 

Share this post


Link to post
Share on other sites
12 hours ago, dleonhardt said:

Distribution on a single machine is only really useful for faster sim times. There are a couple of parts of the FLIP solver which are still single threaded.

This. It can be faster if the synchronization between the slices doesn't take longer than the single threaded parts would take on their own. Doing this doesn't turn 64GB of RAM into 512GB of RAM. All of the slices still need to be running concurrently so they can be synchronized after each timestep. If anything it would use even more RAM running a distributed simulation on one machine compared to a single simulation. Here's some cool reading if this stuff is interesting to you.

https://en.wikipedia.org/wiki/Amdahl's_law

Share this post


Link to post
Share on other sites
On 13/01/2022 at 12:36 AM, lukeiamyourfather said:

This. It can be faster if the synchronization between the slices doesn't take longer than the single threaded parts would take on their own. Doing this doesn't turn 64GB of RAM into 512GB of RAM. All of the slices still need to be running concurrently so they can be synchronized after each timestep. If anything it would use even more RAM running a distributed simulation on one machine compared to a single simulation. Here's some cool reading if this stuff is interesting to you.

https://en.wikipedia.org/wiki/Amdahl's_law

Well that sucks.  So you can not simply sim one slice, finish, sim another etc on one box?  Damn... 

 

Share this post


Link to post
Share on other sites
22 minutes ago, HM_2020 said:

Well that sucks.  So you can not simply sim one slice, finish, sim another etc on one box?  Damn... 

 

No simple way unfortunately. You could do it similarly to PhoenixFD's Cascade Simulation feature, but transferring sim data from one domain to another and have it be relatively seamless, is not trivial to do to say the least.

  • Sad 1

Share this post


Link to post
Share on other sites

What worth to try its make few sims with lower res and combine them, and then mesh result

Share this post


Link to post
Share on other sites
1 hour ago, tamagochy said:

What worth to try its make few sims with lower res and combine them, and then mesh result

Yeah I thought of that, just do a bunch of small domains with edge boundaries, combine all the flip sims into one with soft edges, then do whitewater from that...  but it will look like mixed sims a bit.

 

Share this post


Link to post
Share on other sites

No, you need sim all domain with res that your RAM allow you to sim, but each next sim with different seed, then combine and mesh. 

  • Like 1

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this  

×