Hi everybody, I have a big dream ... why not to build a site where the cfd-code are made available on-line, where by available I mean executable (and the results viewable) directly on-line ? Of course this would be an enormous work to do but I think it should be useful to a lot of people who doesn't have access to a big computer to run a cfd-code. During my career I've writed down a lot of different codes (such as PNS solutor, LES, ...) and of course a lot of other codes could be retrieved from the internet, but an help would be appreciated ! I have already in mind how to structurate this site and other things like: 1) language to use for codes (server-side); 2) software to use for on-line visualization ... What I would like to know is : 1) do You think this is a useful work ? 2) is anybody interested in joining me with this adventure? Thanks for your attention. Bye!

(1). The forum is providing answers at levels higher than the coding. (2). Once you understood the methods and the approaches, it should be fairly easy to write the code. (3). The coding is the necessary step and in many cases, it depends on the person's preference. (4). For example, if you are solving the heat conduction problem, all you need is to specify the method, such as S.O.R. or ADI, and the rest of it can be done easily. On the other hand, if one do not understand S.O.R. or ADI, then it is almost useless to give him the code listing. There is no way he will be able to read the code listing and learn the S.O.R. or ADI method. (5). CFD deals with numerical methods and solution algorithms, the coding part is usually relatively short.

John,

I think the idea here was more about whether its useful to "run and use" a cfd through a web-browser over the internet - not so much about posting source codes themselves online.

I think this idea has potential, and I believe many companies like GE etc are developing their design methodologies around a web-based platform at the front end. Some products like Ansys reportedly already produce html output of results and reports etc.

I suppose to a large extent - this is really just a change to the pre and post processing stages - though it could be a lot of work depending on how good you want it.

It is also likely to be necessary before the ASP model takes off.

Greg

(1). I had mentioned the Internet JAVA CFD approach two years ago. And I was informed that people at Virginia Tech. and MIT were using it in teaching already. (2). For simple problems, I think, it can be done. The only limit is the speed of JAVA and the speed of the user's computer. (3). On the commercial side, I think, I had came across a website where the users can specify the geometry interactively so that the input can be used in the mesh generation (the service provided by the company). (4). For 3-D problems, it is probably easier to download the code and run it off-line. (5). The JAVA wind tunnel can be found in the resources section, I think.

You might find the following link of interest:

http://cfdnet.tuns.ca/

/mark

Thanks everyone for your replies! Of course, John, I'm not interested in simply giving the codes, but rather (as Greg said) to give the opportunity to run these codes. The pure code will certainly be available but this is not my first aim. I do think that the best way of understanding CFD (maybe everything I would say...) is to collect a lot of experience, running and running a code to see how an initial choose influence the results. As I said, the coding language it is not a problem ... (I can assure you that ASP has the ability to implement a complex mathematical code). It is obvious that my target is not a CFD-expert (as a first step), but more realistically a CFD-beginner (students) that maybe are interested in viewing how a CFD simulation works without spending a cent (except the internet costs). So it is clear that the codes available won't be so heavy to run ... Thanks Mark for your suggestion.

(1). There is a danger in showing the pictures on Internet without a systematic treatment of CFD solution. (2). The understanding of the numerical algorithms, the turbulence modeling, and the mesh generation are quite involved. If the solution is not mesh independent, then the solution presented is not correct and will cause permanent damage on the user's understanding of the flow problem. (unless the results presented are analytical solutions) (3). So, regardless of what you do, you must do it right. The solution presented must be the correct solution, not just a beautiful picture. (4). My feeling is, more companies, large or small, will die because of improper use of CFD in the design and analysis. The situation is very difficult, because on one hand they are trying to use CFD to reduce costs, on the other hand, they don't have a single person who really know the CFD process. (5). The bottom line is: if you are asked to solve the flow over a backward facing step, then you must have the capability to write down the equations, select the turbulence model, and write the code and find the mesh independent solution. Even at that point, the solution may not be correct. So, modification to the turbulence model is required, in order to get the solution validated. The real world problem is much more difficult than that, so, CFD applications have created a lot of good pictures instead of good designs. (6). "running a code alone will not generate the correct solution, it will kill you and the company". The CFD approach is very attractive to a manager of engineering, but it is deadly if his experience is just running someone's codes. This is the first-hand information, to know the future of a company (engineering), all you need to know is whether there is a real CFD expert in their design and analysis department. (where is engine company-L? engine company-A? engine company-H?, engine comapny-W?,.....the list is growing.)

Dear John, I do agree with your vision : CFD is a very complex matter, and even the simplest analisys requires a specialist(job for us ..eh..eh..eh!!!). However I think that this is not my (our) problem. I mean, I'd like to give some instruments to improve the approach to CFD ; it is not my problem if someone use them to obtain something that goes beyond their aim. Certainly this site will contain a substantial part concerning CFD theory and every code will be fully described and validated (as far as possible); main issue will be mesh independent solution, convergence ... . Please don't take my words literally : when I say "experience acquired running codes" I don't mean this can avoid from studing theory (equations) but simply integrate a CFD course. Hope you'll agree with me. Bye

(1). I think I can understand your position. I mean, if the project engineer is not reading this forum and then steps into a man hole, then it is his problem not yours. (2). The forum simply provides voices from different directions so that the reader can make his own screening and judgement. (3). I think, interactive computing on Internet is fun and educational, as long as we recognize the limitations which exist in today's technology. So, it is a matter of defining your scope of Internet computing. With that in mind, I think it is a good idea.

Thanks John, I hope you'll visit my site when it'll be ready. I'll post on this forum the date. However I do still need some help, anyone interested please post me an e-mail. Thanks

I think this idea is applicable in the commerical world too - as I mentioned above large companies appear to be building web-based platforms for their design teams - presumably this can help by "centralising" the computer facilities on which the computations run and allow more collaborative design through the internet - the idea that the US guys work their day and the problem gets transferred to India/Europe .... for further work during the US night etc.

I think though that for any "real" problem - the computations are still done on a largish computer by a dedicated code - not Java in the browser - though that might be feasible later.

Greg

Hi greg, you're right, but, you know, "thinks bigger make smaller". What I'm proposing is just a first step, depending on the results we could expand on a more professional field. Don't worry about software technologies;as you say Java is not yet the best tool to develop a numerical code, but I have in mind a software structure (language could be Fortran,C++,VB,... or anyone you like) that allows to run your code (server or client side) using the Web just as an interface. Currently I'm working to develop such interface on a Windows system (... I know ... I know but this is the most used OS); next step will require the development of a graphical tool to visualize results. Bye.

>>CFD deals with numerical methods and solution algorithms, the coding part is usually relatively short<<

But this really depends on the scheme you wish to solve and what you expect your code to do.. I find the coding part sometimes to be quite challenging, and definitely not a level lower than the CFD algorithms understanding. Writing a general code which is dimension-independant for example (that can tackle 1, 2 and 3 dimensions but with the same loops) is not a straightforward thing from a programming point of view, even on a structured mesh. Unstructured meshes with arbitrary cells shape requires, I think, much more work on the programming front than on the CFD algorithm development. Threading a code for SMP machines is another non-trivial issue for certain schemes. Parallelizing it for beowulf clusters is yet another matter.

I believe CFD researchers in general under-estimate the ability needed to write a good, clean and performant code, and should really consider asking an expert in programming to do it for them. Just my opinion anyway..

-bernard

In a certain sense you saw right, but you have to consider that to develop a good code you have to be a good programmer and to develop a good CFD code you have to be a good programmer and a good CFD-man ... which of this two aspect is more important? This is a refrain of a message posted some time ago : If a good CFD engeneer is also a good programmer (...what a long time to obtain such skills!) how high should be his salary ? What a complex matter CFD is !!!! Bye

Several companies are already offering this type of services - there are a few links available at:

http://www.cfd-online.com/Resources/misc.html#online

There are also several new online "cpu-server-farms" about to be launched which will offer CPUs and CFD-codes usable directly online over the internet or over secure networks.

I agree that this is a very exciting development for the future, although I still think that the concept is not mature - it needs faster network and better security/security-awareness (most companies would never allow an engineer to upload their next design to a server on the other side of the world, no matter how secure the engineer says it is.).

(1). I think, the programming part is important. And, in order to make the code general, easy to read and efficient for different applications, it requires a professional programmer. (2). But if the is a single error in the numerical algorithm or turbulence modeling, then even the professional programmer is not going to produce a code which will generate good results. (3). In the last several years, I have seen companies spending a lot of time and programming effort to streamlining the code, the post-processing, the interfaces etc. But, in the end, the only thing the code can do is to produce the wrong results. And my comments was, the improvement actually push the time "backward" instead of forward. Naturally, these codes were abandoned, because it is impossible to "undo" the professional programming based on the wrong methods, formula, or equations. The errors mainly came from the poor CFD knowledge of the professional programmer. (4). I would be very happy if a code can solve just one CFD problem. And we are not there yet. A code to solve many problems (a general code) is almost like a joke. That is why we are here.

Hi Jonas, I've visited sites you proposed but I found some differences between their and mine vision. Someone of them are providing online exec but you have to pay (10 $ per hour) and afterall is not so cheap considering that to obtain a serious CFD result you have to run several times a code that could be very time-expendable [ex: 2 days (very short indeed!!! at least for a professional code.) = 48 h * 10 $ = 480 $ for a single exec !]. Others provide codes for educational purpose but with a scarce integration with theory. Of course I invented nothing ! I agree with you for security problem but this will be a future issue (at least for me!).

Well, finally we're going on a delicate field : you can optimize a code from a programming point of view, but none says the results it gives are right. The main problem is that CFD models are not general by definition, infact the only complete and general way to solve a CFD problem is to solve complete N-S equations but this is still not affordable. In this sense every model you choose is just a (very) particular vision of the complete problem and has a (very) narrow field of application. Moreover new codes develops as the speed and capabilities of the CPUs is increasing (Moore law), so it is clear that once the code is developed (and fully validated) an optimization process would require a too long time, and a new code becomes old before being released. The only strategies we can apply (from a programming point of view)is a fully detailed project from both point of view functional(CFD-a sort of flow chart) and structural (programming). This is usual practice in developing a standard (management) application, not yet in developing a CFD application.

Yes you're right - its happening.

Here in Sydney, a new computer centre housing a IBM SP2 (68 processor) a SGI Origin (64 processor) and NEC-SX-5 was opened last month with the aim of providing some computing power to researchers (like me) but also to sell computer time to companies from all industries - to run their apps - including CFD. (though probably will mainly be banks - datamining their databases to send out junk mail!)

I suppose the aim is only partially commercial, since it is hoped these centres act as an impetus to innovation etc ....

The model at least in part relies on fast secure networks.

Greg

i think it's a great idea. the ability to write CFD code has nothing to do with someone's ability to write code. plus if you want to evaluate the all different schemes out there you wouldn't possibly have time to write all that code must less have time to test all the different schemes. plus why should a CFDer have know the nitty gritty of grid generation and visualization. Nowadays any user of a 2-D unstructured code can easily download anyone of many grid generators. I don't think the user should have to learn the nitty gritty of that (the grid gen code is usually longer than the CFD code itself). as far as CFD schemes go I can say thatI've only actually coded one type of scheme (Jameson type) but i understand many other types. I have encountered a lot of people who know a lot about CFD but have never done any coding. from experience I'd say that coding is a mechanistic and not very intellectually demanding. plus as we've seen many times on this forum people make mistakes coding or haven't had access to the latest information etc (this is particularly true for people in third world countries). to allow users to gain more experience it would be good for the site to have a suite of test cases users could run and compare to the applicable test or analytical results. Also a thick theory section would be useful actually it need not be extremely thick. I learned as much about turbulence from the online theory section of the Flo++ manual. so I really think it's a good idea.