Learning Forum

[jfonken]

Hi all,

For my PhD project, I'm working on FSI simulations of abdominal aortic aneurysms. Most of the challenges I faced have been solved, so now I'm looking into the efficiency of the simulation. I'm running my models on a high performance cluster (HPC) in which you would claim multiple nodes if needed. When you claim a node, you'll get access to all of the 24 processors that the node has.

Since the Fluent and Mechanical problems are solved separately, I would expect that I could use 24 processors for Fluent and 24 processors for Mechanical. When I ran a simulation on my own laptop using 4 processors, I was able to allocate 4 processors for both processes using the following commands (part of my 'runFSI.txt' file):

[Stephen Orlando]

Hi Judith,nRunning on 24 cores for Fluent and Mechanical will require 4 HPC Pack licenses. Can you confirm you have these licenses?nPlease have a look at this section in the documentation in detail, it has everything you're looking for. https://ansyshelp.ansys.com/account/secured?returnurl=/Views/Secured/corp/v202/en/sysc_ug/sysc_userinterfaces_advtasks_parallel.htmlnIf you're using a PBS, LSF, UGE, or SLURM queueing system all you'll need to do is specify the core count for each solver with the PartitionParticipants command in your runFSI.txt. System Coupling will automatically detect the machine names for these queueing systems. n

[jfonken]

Hi Steve,nThanks for your fast reply! In the license status, it says that I'm using 8 licenses at the moment. However, I don't know if these are Pack licenses, so would you think this is enough?nI'm using SLURM, so I will definitely look into the PartitionParticipants command, thanks!n

[Stephen Orlando]

Glad to help. Yep, those are 8 HPC packs, so you're good.n

[jfonken]

Good to hear. When I run my simulation for a longer period of time, the solution does progress. However, it's very slow (in comparison to running with 12 processors for each participants), although I would expect using more processors would speed up the solution. Do you know what could be the cause of this problem? n

[Stephen Orlando]

Hi Judith,nWhen the SharedAllocate* partitioning algorithms are used, the solvers will share memory. If there isn't enough memory, the solvers can run slowly. Keep in mind that the solvers still use memory even when they aren't solving. This could be what's happening.nStephenn