Parallel Computing

[evgin]

Hello

I am going to parallelize the code for my 3D-8-story RC building using OpenSeesMP

There is going to be a vast number of parametric studies in which input parameters would be taken into account as Probability Distribution Function (pdf)

Besides, record-to-record uncertainty in ground motion would be considered

My question :


I am going to follow Frank’s presentation and add that 6 lines of code to make it work in parallel , but I have several thoughts going on in mind as I am new to parallel computing !

1- If I only do the analysis for a suite of ground motion and do the parametric study next, then would not it be better that I do the domain decompression in OpenSeesMP meaning that I assign various elements to different processors? or it is automatically done meaning that my model is run in multiple processors for each ground motion ? Is it automatically decomposed?

2- Using that 6 lines of code, each dynamic analysis is run in a separate processor whose ID is defined by the user using “if {[expr $count % $np] == $pid} { , but if if I wanna do domain decomposition , then should I use a similar if command to define various nodes and associated elements under specified processors?

3- Same for the parametric study in terms of monte-carlo simulations , under a suite of GMs, Can you please advise me on this a bit so as how can I parallelize these types of analysis ?

Thanks
Reza.

[fmk]

which to use depends on how many processors you have access to and how slow is the model running sequentially .. if #cores is limited and way more MC simulations to perform than cores, don't bother trying to partition domain.

2. yes

3. not really possible, you would have to use OpenSees to start other OpenSees applications.