[dgould]

Simple theoretical question: It appears to me that most operations in Ansys CFD post (and pre, for that matter) are only single threaded. In most cases, I'm sure there are very good reasons for this that are way over my head. However, when working with time series data to generate, say, a 1-D chart of a spatial average temperature over time, it still appears to be only using one core. Isn't this one particular situation a simple SIMD case where is should be ridiculously easy to parallelize? I ask because my programming skills are crap, and yet even I know enough to always do such when I am doing data analysis in Matlab. Thus I am assuming that I don't understand something about the process here, and that's quite frustrating. Any ideas?

[ghorrocks]

The term is actually embarrassingly parallel https://en.wikipedia.org/wiki/Embarrassingly_parallel

Yes, what you describe is embarrassingly parallel. But you still have to write the application to split it up, send it to multiple processes and gather the results together at the end. CFD-Post has been written as a single threaded application from the start so I suspect that the code base won't support this sort of thing. Also the majority of the tasks done on CFD-Post are not easily multi-threaded, so the effort spent making this task multi-threaded is not going to make any difference to most users, so the effort does not have much reward.

CFD-Post, is a post processing package of moderate power. If you want to do multi-threaded post processing you should consider high-end post processors like Ensight, Fieldview or Tecplot.

[Lance]

Originally Posted by ghorrocks

CFD-Post, is a post processing package of moderate power. If you want to do multi-threaded post processing you should consider high-end post processors like Ensight, Fieldview or Tecplot.

Or ParaView. It can be quite tempremental, but I use it all the time. And it's free...

[dgould]

Well, I would use Visit, but I haven't been able to figure out how to efficiently export the data in cgns for it. Or, at least, no efficient way. The only way I found appeared to be single threaded as well as was slower than dirt. But I could have been screwing something up. How long is exporting a run with, say, 1E6 points, supposed to take?

[ghorrocks]

1e6 points where each data point is a single value will finish in a flash. 1e6 points where each point is a full save of the simulation variables at that time will crash most post processors. So it depends on what you are doing.

If you are doing seriously large post processing then you need to invest in a seriously powerful post processor. I do not know visit, but the one's I listed (and Paraview) are definitely powerful and able to handle much bigger data sets than CFD-Post.