I think Charles summed it all up. For academic world OpenFOAM is indeed much better software than other commercial CFD codes. As a research student i am not aware of problems that you guys face in industry. Yes, OpenFOAM has its drawbacks like CAD, meshing etc but at the end of day its absolutely free software and it is improving rapidly. Therefore i am of the opinion that OpenFOAM will eventually take over commerical CFD codes. When OpenFOAM is future then it is better for a research student like myself to stick with it.

Marco

>>Yes, OpenFOAM has its drawbacks like CAD, meshing etc but at the end of day its absolutely free software and it is improving rapidly. Therefore i am of the opinion that OpenFOAM will eventually take over commerical CFD codes.

That is some logic - OpenFOAM is improving rapidly while commercial codes are freezing up like ice. I think your position as a graduate student shades your views.

W.

As a research student, if you can not get help from your pioneers in the use of openfoam, I think you never try it. It just wastes your time. Please note that almost all techniques in openfaom has been well developed to some extent. What it is your originality in your research? You may say I focus the process not modelling. In this case, why not directly use commercial packages. you may gain more if your time for the study of openfoam has been utilized in your research. However, in a long-term view, your research group can try to replace commercial packages with openfoam. In one word, try to work on openfoam together with others for research purpose, otherwise just forget it.

"are you sure? googling on this issue does not give a sharp judge !"

That's just microsoft marketing and has little to do with the truth. If the microsoft OS is so good why is it that all high end workstations and supercomputers use linux or some other flavour of unix? Actually the google servers all run linux (as do amazon).

Although I agree with you Tom (regarding win vs Linux), that type of "logic" is somewhat dangerous. You can just as easily turn the argument around and say that because XX% use windows it is superior to Linux.

Similar logic:

"Hundred billion flies can not be wrong - eat shit."

=)

It's a good idea actually to test the CFD performance before making big claims about differences in performance between Linux & Windows. There are some good reasons to use Linux for CFD, such as:

Cheaper and easier to set up a cluster

64-bit Linux is now very mature and quite well-supported in terms of drivers, etc., compared to 64-bit Windows XP, where there is an astonishing lack of suitable drivers for some hardware

64-bit Windows versions of most commercial CFD codes only came out several years after the 64-bit Linux versions.

Some software, such as OpenFOAM, won't run under Windows. Other useful software, such as Rhino, Multisurf and Excel, won't run under Linux. (and no, OpenOffice is a very good alternative to the other elements of MS Office, but it is not an adequate replacement for Excel) Most serious CFD practitioners that I know run a mixture of Windows and Linux machines.

However, when you actually measure the difference in performance, it mostly comes down to the particular compiler, not operating system. Code compiled with gcc, for example, is slower than code compiled with Intel or Pathscale. In fact, I have tested code compiled with gcc, and the executions speed was the same under Windows, Linux or Wine under Linux! So it didn't care about the O/S, it only knew about the low-level instructions and the hardware.

"However, when you actually measure the difference in performance, it mostly comes down to the particular compiler, not operating system. Code compiled with gcc, for example, is slower than code compiled with Intel or Pathscale. In fact, I have tested code compiled with gcc, and the executions speed was the same under Windows, Linux or Wine under Linux! So it didn't care about the O/S, it only knew about the low-level instructions and the hardware."

I suspect the computer was not been used for anything else during the test - so the OS did not have to interfere with the running of the code. If for example the job was running in the background while you were doing something else such as editing text in openoffice would this still be the case?

Some points you missed off of your list include stability (linux rarely needs rebooting) and the fact that it has had support for multiple cores longer than windows.

Tom,

i'm not biased to win or linux, currently i use dual-boot system.

question to your answer: linux is used almost in cluster technology becasue until 2003, windows does not support needed message passing like MPI, but after 2003, server version of windows is appeared and to my knowledge (related to 2006) it successfuly manages more than 256 nodes with comparable performance andscalability. Today when i enquiry aboyt its, i read something about 1000-10000 nodes!

So in future it be possible to compare.

Another critical issue (i select linux just for this reason), windows does not support diskless cluster, i.e., each node should has a hard-disk and a distict registered windows (expensive: hardware/software, important for large cluster).

In my experience with both system, i could not say what is better, because contrast between performances is not sensible (i did not do a profiling sofar).

Finally word: usually scientific community reject windows by default (is somehow emulation), but if we see current sitution, you see that windows is very common and public, e.g., read 2007 static report of CFD-Online Forum which show windows be dominant OS !

"I suspect the computer was not been used for anything else during the test - so the OS did not have to interfere with the running of the code. If for example the job was running in the background while you were doing something else such as editing text in openoffice would this still be the case?"

Pretty much. CFD uses so much more of the CPU than anything else that normal "office" tasks have very little influence. Sure, if you are doing some interactive things that require a lot of CPU (like CAD modeling), you will notice the effect on the CFD performance. But this is determined by the relative priority that you assign the two tasks, something which is equally easily done under Windows and Linux.

"Some points you missed off of your list include stability (linux rarely needs rebooting) and the fact that it has had support for multiple cores longer than windows."

XP Professional is pretty stable, you know. It doesn't need many reboots either. In a way I'm playing devil's advocate here. I too prefer Linux for a lot of my work, but the fact is that it doesn't hold a huge advantage at the moment. Where Linux has generally had a big advantage is during periods of transition of the hardware. When hardware moved to 32-bit, Linux was a initially a far better multi-tasking O/S, and it was only NT and finally Windows 2000 that made Windows competitive again. Similarly, when the 64-bit move came, once again Linux64 was almost immediately a viable option, but it took a surprisingly long time for 64-bit Windows to become a viable alternative. Even now, there are devices and software that aren't compatible with XP64. Anyway, this has gone rather a long way off topic, except perhaps to point out that open-source evangelists need to keep an eye on reality as well!

Charies, fine idea, it is indeed correct.

e.g. gcc could be run under cygwin or mingw in windows which is slowdown due to this (i compare with MSVC 6. and show 40 % slow down for my application).

so if anyone compare a compiled code by GCC in windows and linux, his comparison is not fair !

So mean of you walter! You also started work or research (or whatever you do) as a graduate student. Didnt you?

Thank you for your reply chen.

Marco

"question to your answer: linux is used almost in cluster technology becasue until 2003, windows does not support needed message passing like MPI, but after 2003, server version of windows is appeared and to my knowledge (related to 2006) it successfuly manages more than 256 nodes with comparable performance andscalability. Today when i enquiry aboyt its, i read something about 1000-10000 nodes!"

As far as I am aware none of the major "supercomputer" manufactures use windows and so I suspect that the "1000-10000 nodes" comment has nothing to do with a full scale MPP fluids simulation.

Everyone knows, rightly or wrongly, that windows is the dominant OS for business and home use. It's not so clear for scientific computing - Sun, HP, Cray, NEC, IBM, ... all tend to use a version on Unix in their systems.

> I suspect that the "1000-10000 nodes" comment has nothing to do with a full scale MPP fluids simulation.

I did not use myself and as i said i only read about. look at this link (1024 node is claimed to me commersialized):

http://h18004.www1.hp.com/products/q.../12574_na.html

this presentation gives interesting results:

http://www.winhpc.org/dl/presentatio...rd_HP_WCCS.ppt

comparison between Linux/windows cluster, it showed that difference is -+ 10 % (somethimes win is supperior and sometimes linux !),

particularly it has some results for runnig Fluent on both system, which could be interesting to this Forum's member (in this case Linux is 5-10 % better), note that it is not sign for superiority of Linux becasue all thing are not same (at least programming of Fluent and its used compiler for each OS)

PS: I priori know that some peoples with closed eyes still say Linux Linux ... but we do not say any becasue fair comparison is not simple

One thing is a benchmark and something completely different is the daily use such a configuration

Ok bottom line is OpenFOAM is better. Thank you all for participating in the discussion.

Marco

Who said that? This is what your original post asserted. There were various arguments either way..now you "close the discussion by repeating your statement". I give you an F in debate.

These arguments are always very silly and never get anywhere.

The real "bottom line" is that each is a tool. In the correct hands each tool can be made to do some nice work. Similarly in the wrong hands you can just as easily produce cra* answers with any of the codes.

The real bottom line is you have difficulty in understanding English. It has been clearly written by many posters that OpenFoam is nowhere near to commercial solvers in terms of features etc. It is good program but still got to be developed a lot.

You can live with your bottom and line though. No probs.

Apart from air and possibly water, what free thing is better than it's commercial alternative? OK, OK, apart from price...

One of the advantages of a capitalistic culture is that if it's good someone will want to make money out of it and will do so if people want to buy it. Personally I like to pay for things, a certain amount of responsibility is then vested with the seller to honour the price they are charging. Without this albeit profit driven relationship, one has to rely on trust and integrity. Honourable goals indeed but ones that tend to be corrupted by our rather Darwinistic approach to commerce.

Maybe I missed it in this thread but how does OpenFOAM Ltd. survive? Is it beggarware? Does it rely on advertising? Does it make it via consultancy services? I'd trust it more if I knew that, like the rest of us, it was trying to make an (honest) crust!

The very fact that "opaque" does get so upset about the post shows that Ansys and others are not taking developments like OpenFOAM lightly ...

I have been personally involved in a comparison between OpenFOAM and CFD-ACE+ for sloshing simulations ... yes, OpenFOAM does not have a shiny GUI and might even be akward to use sometimes, but in terms of performance and accuracy it had an edge over the commercial code. Admittedly that was just one commercial code, and a minor one as our dear Ansys disciple pointed out.

...having said that I suppose open source software development could herald fundamental changes to the way software commerce works. A developer I work with has pointed out that Qt (amongst others) has a model whereby they offer two versions, an open version and one for commercial distribution. The latter making money, the former widely adopted, used and more importantly tested. Loads of web based product provisions are advertising funded. One thing is for sure, one can not assume that the standard annual lease or perpetual+X% maintenance model common in the commercial CFD market will survive intact indefinitely. Is this the start of the end or the start of the beginning?