Learning Forum

[liweiyi]

Hi, 

We are testing the scalability of ANSYS Fluent software. We considered a relatively large resolution simulation (256^3 control volumes, open-channel flow LES) and programmatically increased the number of computing cores. It seems it does not scale well. Attached please find the results of the scaling test.  

 

The first three points correspond to a single compute node (shared memory), whereas as the remaining points are using 2 and 4 compute nodes (distributed memory). What’s surprising is that the time increases as the number of cores is increased, which is the opposite of what we were expecting.   

 

Does anyone have some idea on the possible reason why Fluent does not scale on the cluster?  

 

Thanks,

Weiyi  

 

[DrAmine]

Communication across the compute nodes can be an issue here. Can you add details about the cluster and connection between the nodes

[liweiyi]

Hi, 

The cluster I'm using is Terremoto from Columbia University. The detailed technical information is here: https://confluence.columbia.edu/confluence/display/rcs/Terremoto+-+Technical+Information

 

Weiyi

[DrAmine]

Moving to installation to get better audience as it is not a Fluid question but performance.