CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   FLUENT (https://www.cfd-online.com/Forums/fluent/)
-   -   Fluent speed (https://www.cfd-online.com/Forums/fluent/35266-fluent-speed.html)

ahourri December 13, 2004 10:04

Fluent speed
 
I am running a very simple application on 9 nodes of parallel pentium-4 computers. The grid I have has around 10 000 elements. It is an unsteady laminar incompressible gas flow. It's taking me now over two 3 weeks to get the desired results while exactely similar problem using pheonix on computer with 2 pentium 4 processors takes about two days. Can anybody tell me if Fleunt can be faster. Appreciated

ap December 13, 2004 10:12

Re: Fluent speed
 
Your model seems very small to be running on 9 nodes - it may be taking so long because of the speed of the computers talking together - try running the model using serial fluent or on less nodes

ahourri December 13, 2004 11:00

Re: Fluent speed
 
Thanks for the reply, but I tried using less nodes (3 nodes).Still running slow

Jason December 13, 2004 11:20

Re: Fluent speed
 
10,000 elements should be able to run fine on 1 processor (and they say 600,000 to 800,000 nodes per 1Gb of RAM, so you won't be using much RAM). I agree with ap that your network chatter is probably overpowering your computing power. It could be chosing the wrong MPI for your network setup as well and slowing you down further. Parrallel processing is touchy, and if you don't absolutely have to, I wouldn't.

I recommend trying 1 processor, especially if they're P4s. I run 100,000+ on a single 500Mhz UNIX machine with pretty good speed.

Of course, I'm assuming the problem is the amount of time per iteration. If this isn't what's going on, then post some more info.

Hope this helps.

Jason

ahourri December 13, 2004 12:18

Re: Fluent speed
 
thanx jason, I will ask our system administrator about The MPI used.

ap December 13, 2004 17:37

Re: Fluent speed
 
Hum, it seems like someone is using my nick! I didn't give this answer!!

Hi,

ap :-D

ahourri December 15, 2004 14:58

Re: Fluent speed
 
Thanks again Jason, you put a finger on the problem, we use socket as an MPI, but the error we did, is that we setted the process number to 1.This number is now setted to zero and the simulation is pretty fast using only one computer. We use SOCKET because it allows us to distribut our work on given nodes as we have many applications running. Thanx again

Jason December 15, 2004 15:28

Re: Fluent speed
 
Anytime I can help

:)

Jason


All times are GMT -4. The time now is 00:44.