Hello to the community. I'm looking for advice on getting started with a basic CFD capability at home (small simple models, Euler would be fine). I would prefer not to spend much on this experiment at least not initially. I'm not completely uninitiated, but it's been a long time. I haven't used Unix or Linux in a while, I've been living in a Windows world. I did my MS in fluids in 1996 and wrote a 2-d ENO Euler solver from scratch that was based on a Godunov-like scheme with a Reimann solver with characteristic sub/supersonic BCs in FORTRAN. I made grids for simple geometry by writing short programs. I thought that was interesting because of the apparent analogy to little shock tubes.
At work, I've had occasion to make only one or two models like a 3-d NS NASTD aircraft model (ICEM grid 4-5 million points) with flowing inlets, but always with help from company CFD colleagues. Sadly, I have not had the need to use CFD in my work for the most part, we have a CFD department.
Dare I ask, is there a recommended free CFD SW package? I may need some hobbies soon, so setting this up could be one.
I have a P4 2.75 mhz pc sitting here on which I could load Linux for the inital test before buying more capability.
Thanks in advance for your advice.
Im not sure what the "SW" stand for in "CFD SW package" but i am currently learing OpenFoam1.6. Some people who have used other free cfd packages tell me that its code is laid out weird, but since it is my first opensource cfd package it works great for me.
Id recommend visiting Madds' Cookbook for an easy installation:
I was using a 2.8ghz and just upgraded to an i7. I did some benchmarking and posted it here:
Eric, thanks for your advice. I'm replying via iPhone but when I get home I'll look more carfully at the link you provided. Open Foam was definately one of my best candidates. I'm hopeful that I'll finally get started with some basic capability in cfd at home. Thanks again, and I'll let you know how it goes.
I am told that Code Saturne (http://research.edf.com/the-edf-offe...ne-107008.html) is also very nice.
I went the OpenFOAM route first. Using a 3 mhz P4 and, ok gasp 512M memory, Umbutu 9.10, I attempted to Makewall. I was copying the log to check on the progress and it was still moving along after about 40 hours.I had to leave the house for a few days and didn't want to leave the pc cranking away with an OS I was not familiar with, so I interrupted the make. I think i need to see what obj modules I have and modify the make to do the rest. Do you think this will work?
I am not exactly sure what you are doing, but 40hrs on a 3ghz is pretty extensive. Perhaps you need more memory for your application. Once you go beyond the memory limit of your application it frequently accesses the hard drive turning a 2hour project into a 2week project.
Not sure that answered your question, but its a thought
Eric, thanks, memory did the trick. I could see the swap space was in use before the RAM change. Umbutu's system monitor allowed me to watch memory used (300-400 MB) and swap space in use (100-200MB). These numbers were pretty dynamic so it did not seem like it was that much of a burden on the system to be using that much swap - plus, this was the system I was stuck with (at least for that time).
After considering buying a new computer, and not knowing if any of this would work out in the end, I decided to go the cheapest route and buy some memory. I got a packaged pair of 1G modules, abt $65 (after rebate!) for the two slots I have, max-ing out the memory at 2G on this Dimension 3000 P4 3Ghz. Finding myself once again a student of Linux, I decided to just re-run the full Allwmake step after installing the 2GB, instead of trying to figure out how to salvage what had already been done. I'm not sure what options are available in the make, one must be to skip a step that already has a "current" obj.
With the new memory, the difference was clear immediately, there was no swap activity and during peaks, about 520MB memory was in use. This confirms that 512MB was below the minimum memory for this task. The Umbutu system monitor was showing a plot for two CPUs (is this what hyperthreading is?) and fairly long stretches of 100% on one cpu while the other was relatively idle and then it would trade places with the "other CPU" ( for thermal?). I set out on the bicycle and left it for about 2 hrs. When i returned I was not able to get the screen to come back on regardless of what i did with mouse or keyboard. I had disabled the screen saver and am using a KVM switch which I'm guessing is probably not supported completely by Umbutu. I was looking at a black screen. After a few minutes I did a hard power down. The last entry i the log was
make: Leaving directory `/home/steve/OpenFOAM/OpenFOAM-1.6.x/applications/utilities/thermophysical'
I think it was not done, but had finished enough such that the installation check had no critical invalids.
I went looking for a super-simple, beginning-to-end, check case script. No luck on that one, so this is going to be a veritable plethora of learning for me.
I ran something I found in the tutorial directory, potential, cylinder. It seemed to run ok - less than 30 seconds clock time. I haven't found the easy one-button plotting feature yet (ha ha).
Any suggestions for my next step in the learning process?
Thanks for your advice.
Good to hear that the memory helped.
Yes the 2 cpus is due to hyperthreading. Mine shows an oscillation in 100% loads from processor to processor as well. I belive this is how the software/os perceives it rather than what is actually occurring. It might be interesting to turn hyperthreading off in the bios and see if you get a performance increase or decrease(let us know how it goes if you decide to).
I belive the blank screen its due to a screen lock. If it happens again just type your user(or possibly root) password and hit enter. Sometimes it takes two times for it two work( I arrived at that solution by dumb luck).
As far as a good next step, you may want to try out the airfoil2d tutorial. I put together a simple benchmark forum on it here:
Best of Luck!
You can set your Ubuntu to use the swap only after 80-90% of your physical memory is in use, "ubuntu swappiness on Google".
By default this is set at about 50% if I remember correctly which is stupid if you have for example 2GB of RAM the system will use the swap even if you have 1GB free.
Also you should disable all the graphical effects from Ubuntu - you will recover some extra RAM.
I need a CFD Analysis
Hi, I am new to this forum and to CFD, I need to find someone that can do a CFD analysis inexpensively. Is there a site to go to or any advice?
Great tips, thanks. It should be helpful in the future also.
I just got back from another trip - I hope I can find some more time now to get a little farther with the set-up here.
I recommend that you post your job on this website(cfd-online). http://www.cfd-online.com/Jobs/
There are many people who check this on a regular basis for full time and contract work(myself included).
Good luck in your hunt!
Below is from the log of the run. The time was 57.86 seconds. It's one of the older, coal-fired PCs.
It looks like the extra memory (or slightly faster clock rate) made only a modest difference .
Dell Dimension 3000 Pentium 4 running at 3.0 GHz, single processor, 2 GB DDR RAM.
[installed Patriot 2GB (2x1GB kit) "PSD2G400K" "PC3200 400 MHz"]
Ubuntu 9.10 32 bit
Open FOAM 1.6
But I still haven't found where the pretty colour pictures come out. : )
Do we have anything in the way of grid generation and solution visualisation tools?
Time = 500
smoothSolver: Solving for Ux, Initial residual = 1.08774e-07, Final residual = 1.05461e-08, No Iterations 2
smoothSolver: Solving for Uy, Initial residual = 4.31634e-08, Final residual = 9.44679e-09, No Iterations 2
GAMG: Solving for p, Initial residual = 9.7231e-07, Final residual = 9.7231e-07, No Iterations 0
time step continuity errors : sum local = 1.64311e-09, global = 1.98362e-17, cumulative = 1.2383e-14
smoothSolver: Solving for nuTilda, Initial residual = 2.16267e-07, Final residual = 2.02589e-08, No Iterations 2
ExecutionTime = 57.86 s ClockTime = 58 s
For visualization, pretty much everyone uses paraview. OpenFOAM 1.6 comes packaged with paraview.(lots of info in the user manual chapter 6)
Try running "paraFoam" in your case directory( this is a handy pre-installed function that converts your case to vtk and opens it in paraview)
If it doesnt work you can also use "foamToVTK"; then manually open the .vtk file with paraview.
I have been having good luck using Gmsh for grid generation. I recommend giving it a try.
|All times are GMT -4. The time now is 16:33.|