I've noticed throughout this discussion list many complaints about the cost of commercial cfd software. I was wondering: 1) How many rely on self-made codes to commericial codes? 2) Are self made codes (freeware/shareware/opensource) as reliable as commercial cfd software? 3) Is research being done using self-made codes exclusively?

(1). The best way to die is to use the commercial code. (2). But if you are paid to use commercial codes so that they can die faster, then there is nothing wrong to use the commercial code. (3). Based on my paid job experience, the use of commercial codes is to cover up the trouble of the company from the public, and most of the time bad results were covered up so that good picture can be painted to upper management or the public. (4). To know whether the company is in trouble, just ask who is doing the cfd work. The professional has his signature in his professional field. (5). You can't say the since everyone is polluting the air, so the conclusion is it is alright to do so. (6). Using the commercial codes is all right, because it is the intermediate solution. But if one use commercial codes as long term solution, he can not compete with others and eventually will die. (7). So, the use of commercial codes is to by additional time, so that long term solution can be found. As long as this is understood, it is not important to know whether a code is commercial or not. The solution is not in the codes, and it is essential to find the right person to solve the problem. (8).The codes he is going to use to solve the problem depends on the problem and the person. Since the problem can not find the right person, someone has to find the right person to find the right solution to the problem. If the right person can find the right solution, then the methods and codes he uses are not important at all.

Actually, I was wondering how extensive non-commercial CFD software is being used to solve real-world CFD problems.

(1). I can give you a hint. (2).A few years back, when I was working for a company over 50 years old, they have several hundred computer codes on the system for the product development. About half of that number were used on the routine basis. That means over a hundred. Aero-related codes must be several dozens. (3). There were a couple of commercial CFD codes on the system, being used to handle some non-standard applications. In most cases, the results could not match the test data. (4). This is only one example, but it is fairly typical. Most large and old companies have their codes for their product design and analysis. But this does not include all types of company.

I think self-made codes somewhat specifically purposed. The use of self-made codes in real CFD world is limited. You can use your own code well in the problem you have devoted. However people who use your codes to solve another problem will encounter many troubles. It is due to some bugs in your code. These bugs are not found in your application. They become active in other applications. They may come from many aspects such as incorrect mathmetical treatments for the physical phenomenon, unexpected singularities of determined coefficients, and improper matrix solver, etc. It takes time to modify a self-made code when these codes are used for different purposes. Many modifying and test must be done before the self-made codes become generally availible. All above is just some of my opinions.

Personally, If I could get a commercial code to work for my research I would use it, because they usually come with all sorts of goodies. But unfortunately when we use a commercial code it crashes. Every time. The change needed to make in the commercial code is simple but without the source code is practically impossible.

Can the beowulf cluster support CFD usefully vs. SMP computers?

Does anyone know of a good non-commercial CFD software package designed to run on a beowulf-style cluster?

The answer to your first question is yes.

the answer to the second is yes but I couldn't say that they'd necessarily work very well. Reviews I've gotten are mixed. Most commercial parallel codes prefer SMP machines I think.

I totally agree!!

In My department, most of the engineers work with commercial codes. But for very specific jobs, they may used home made code.

From my point of view , commercial codes have one big advantage, they offer an overall solution: mesh generator, NS-Solver, and post-traitment. If you use a home made code, you have to find a software to generate your grid and another to do your post-traitment.

From the small experience I have with the industry, I can tell you that they prefer commercial codes. And not just one, depending on their needs, they may pay for two or three different softwares. For very specific jobs, they may ask a research group to develop a code.

John,

I think that the answer to your question depends on what you are trying to do. You really discuss three kinds of software: Commercial, Existing Freeware, and Self-Developed. The things that will drive you to a decision are: problem complexity, problem repitition, and schedule.

If you have a very specialized and complex problem, then you may have trouble finding any existing code that will solve it. You may now have a choice other than writing your own code to solve it. If you do use another type of code (commercial or freeware) then you will need to modify it to do what you require. Commercial codes usually allow the user to develop specialized subroutines and link them with the main body of the code. With freeware, you might have to tinker with the guts of the code, and you would have to do it without the benefit of user support.

If you only need to solve a very specific class of problem, then it may be best to write your own code, or get a freeware code that is good for that type of problem. Generally speaking, purpose developed codes get better answers on difficult problems than more general codes. This is simply because the developer knows the "correct" answer based on test data, and tweaks the code until it produces that answer. Commercial codes often include comprimises that don't necessarily make the answers wrong, but can cost accuracy in specific situations. The choice of turbulence model can make a significant difference in some situations, for example.

The final criteria is time. If I had time, I'd build a 1/4th scale replica of the Eiffel Tower in my back yard, but I don't. (That and my wife wouldn't appreciate loosing all that empty space behind the house.) A good engineer with a solid background in fluid mechanics, some modeling experience, and effective training can become proficient with a commercial CFD code in about 1 month. A moron will always be a moron. Developing, testing and tweaking your own code would require (I'm estimating here) at least 6 months. So it depends on how long you've got to devote to the effort. If you need a design in a couple of months, then you need to become proficient with a tool in a short period of time. If you're just taking your first class toward a Ph. D., and you've got 5 years, then you might as well start coding.

Regards, Alton

(1). Based on my experience, many companies failed because their products can not compete with other companies' products, for many reasons. (let's rule out MS.) (2). These companies all have their products on the market, so they have all solved their simple problems. They don't need a general code to solve their simple problems, that's what I am saying. (3). So, they are hoping that CFD can solve their complicated problems to give them some edge to win the competition. In other words, a large effort and small gain. (4). If the general code can not produce that small gain needed to win the competition, then is the faster solution still viable to keep the company in business? (5). I am not against the commercial codes. But if one-month training of an engineer is enough to take away jobs from a PhD, then they should shut down the school in the first place, include George Bush's plan to improve the educational system as well?

John,

It depends on what you define as a simple problem. I'll give you an example:

I recently did some work for a friend of mine who works for a company that manufactures relief valves. They won a contract to provide a vacuum relief valve that would pass a certain flow rate of air at a fixed differential pressure.

As CFD problems go, it was not difficult, but as a design problem it caused them a great deal of trouble. They started out by designing the valve using tried and true handbook equations, and some scaling from a valve they had built 20 years ago (to different requirements). They did the engineering, made drawings, forged a valve body, machined and finally assembled a valve. Then they set it up in their test facility, set up the required differential pressure, and measured the flow rate. The result they got was much lower than they needed. What they had was a $30000 hunk of scrap and no firm idea of what to try next. That's why they came to me.

I treated the valve as axisymmetric (not very much of an assumption), and solved flow and turbulence. I was able to generate a mixed quad/tri grid in a day, and could run a solution to convergence in a few hours on my desktop computer. The flow rate I got was off by about 5%, but more important to them was that they could see that the valve was poorly shaped and why. Armed with that knowledge they redesigned the inside of the valve and I repeated the analysis. The computed flow rate was higher than they required, so they built a new valve. When the re-designed valve was tested the flow rate was within 3% of the computed value.

The moral of the story is that simple is a relative thing. They thought CFD analysis was complex, and their old analytical methods were easy. I looked at their situation and told (and then proved) that CFD is easy.

As for what people with Ph.D.s should do, the answer is obvious: go to work for commercial CFD companies developing new and better tool to put in the hands of the end-user.

Alton

(1). Very good. I think, you have just outlined how the CFD should work. (2). If the valve company had decided to send their engineers for one-month intensive training, what would be the chances of getting the good design by their engineers? (3). It seems to me that the key to the good design was the interpretation of the not very accurate cfd flow field results. So, I would say that it was likely a team guessing work. And it is equally possible that a person with extensive knowledge of fluid mechanics and flow field could also come up with a good design by just looking at the old design. (4). And if you give the old flow field and the new flow field to an engineer who had only one-month of training and without prior knowledge of the design problem, will he be able to pick the right flow field from the two results? (5). I think, we are telling the same story. And your story has everything we need for how to use CFD. (you are not just an engineer, you are one of the very few persons who read and write in cfd forum. ) And I have to say that it is a success story of "your" CFD application experience. You and I should be excluded from the examples.

Alton, It is nice to read that success does occur in CFD.Did you code the valve or use a commercial based code.That simple word -ASSUMPTION also can make striking differences and it crops up a lot in CFD.

A code for all known flow conditions-all the computing strength you can muster -but will the experts agree on the end results and moreso, will the product benefit hugely after a year or two in service.

I'm glad some people still know how to apply (commercial) CFD codes for industry.

What you're demonstrating is that CFD is not as mature as other simulation tools. Who would even think of building a scale model for routine stress calculations these days? Furthermore, who would write their own solver for doing such calculations rather than taking one off the shelf?

Given time, CFD will travel down the same de-skilling route.

(1). Even today, the structure failure is still not uncommon in new design of combustor, compressor blade, turbine blade, simple radial turbine volute, aircraft control surfaces, launch vehicle, rocket motors, etc...even with off-the-shelf finite element code analysis completed. (2). Failure normally is not printed in the company's newsletter. They don't want you to talk about it. It is bad for the company image and stock price. (3). It is still too early to say that FEM structure analysis is mature.

"Who would even think of building a scale model for routine stress calculations these days?"

In the current Mechanical Engineering, there's a report on testing of a scale model of the pressure vessel that was damaged in the infamous Three Mile Island nuclear reactor accident. The tests were/are being done at Sandia National Labs, a US Department of Energy facility, but the interest and support are international.

I guess the key is that the baby was and is not routine!

You miss the point.

FEM tends to be used upfront, with mechanical testing used to understand failures. With more immature (less trusted) technologies like CFD, it's usually the other way round - CFD is used to investigate odd behaviours spotted in physical tests.

My point is that CFD codes are generally run by CFD people, who need to understand how the codes work to use them properly (at all even?). This is not the case with more mature codes (maybe "spreadsheet" would have been a better "mature code" example - I haven't bothered writing one of those since about '87).