Hello all

I am currently testing out a new cluster that we are going to acquire soon. I am trying to run Fluent jobs (5000 iterations) first on a single node (8 processors) and then on multiple nodes (16 - 48 processors). The problem is, I have no way of knowing how well the case is spanning across these nodes. In other words, I need to record the run-times for each if the runs ( on 1 processor, then on 2, thewn on 8, then on 16 and then on 32 so on...) because I believe that these times will give me an idea of how well the run is getting distributed on the nodes and how the run-times compare against one another based on the number of processors.

I asked a Fluent Tech support person the same thing and she said she does not know of a solution for this, which I thought was unbelievable since most of us run our Fluent jobs on clusters and there must be some way to compare the run-times. Right?? I know that when the Fluent run starts, it shows a corrsponding time beside the iteration number, but that is not very accurate.

Any help is greatly appreciated.

Ravi

I would use either Matlab or C++ to launch all of the Fluent runs and the use either tic/toc(Matlab) or stopwatch(C++) to time each of them.

So there is no simple solution to this like including a flag or something when I launch the Fluent run?? The problem with Matlab is that I have to make people buy it and then convince them to install it on the cluster, which is not exactly pain-free.

If you write the code in C++ using stopwatch it should require a few lines of code and should be pretty simple.

http://osr600doc.sco.com/en/SDK_c++/...ution_Tim.html