open TELEMAC-MASCARET The mathematically superior suite of solvers

C:\opentelemac\v6p1\validation\telemac2d\tel2d_v6p1\NewLemSim\FlowSim.tif
Dear All,
I am trying to run the parallel version of Telemac both 2d and 3d, I am getting a very good speed up for the 2D case. I am simulating a 5 day flow scenario through a river-confluence (attched-tif).

When I run the simulation with the serial version on windows 5 day flow takes 26 hrs on the machine to finish.

When I run the same on linux machine with 8 procs, this is done in 3 hrs 20 minutes. Very happy with the speed up. I have 20,485 points and 35,755 elements and when I go in 3D I use 9 layers.

My question is, has any one got any idea how will the speed up be in the 3D case, my first trials are not fecthing me any significant speed up on 8 procs, 5 days scenario is taking close to 3.5 days to finish. I am also attaching the cas file for 3d simulation, has any one got any idea to speed up the 3d simulations.

Any and all help is greatly appreciated.

Hello

If you have an issues with the picture, please can you send it to the forum support email address: I will add it to your message.

Thank you

Dear All,

Once again if some body can take a look at my case file and point out what am I doing wrong, I will really appreciate.

The simulation in this case is super slow and seems to get even slower after just first 3 iterations.

Thanks,
Sumit

Hello,

Except that the keyword PARALLEL PROCESSORS=4 is commented out, I see no reason why you don't have a speed up in 3D, we generally get about the same speed-up as the same case in 2D.

JMH

Dear Jean-Michel,

Thanks for your reply. Let me ask you two more qiestions.

a) Am I right in thinking that hydrostatic version should run faster than the non-hydrostatic mode, I do realize that the decision to use one or the other depends on the kind of problem you are solving.

b) Would sigma mesh in vertical be any faster than the homogeneous mesh. I ask because the case that I am simulating, I have also used EFDC (EPA model in US) on the same problem and in that model, equations are in orthogonal-curvilinear coordinates in horizontal and sigma mesh in vertical with hydrostatic pressure assumptions. I want to use Telemac-3D because of the unstructured mesh and mpi-parallelization. We have parallelized EFDC with Open-MP paradigm.

Thanks,
Sumit