Folding Forum

Here is my fantasy client, this is not a complaint:

1) All FAH data storage for all machines located on one machine. Z:/miscdata/fah/machine001/SMP/, Z:/miscdata/fah/machine006/GPU, etc.
2) Allows on/off timer of all machines from one application.
3) Allows WU duration cap: "Machine 003 - No WU's over 12hrs." "Machine 008 - No WU's over 24hrs."
4) Sends email out based on switches - "Alert - Machine 005 will not complete WU before requested time (08:00)", etc.
5) Reports estimated FAH gigaflops for each machine and total. (Fun is nice too)
6) Reports estimated power usage per machine and total.

1) You mean that all F@H Data is stored on a "F@H Server" (one of your system) while all the other machines are "F@H Clients" (remaining systems)? An interesting idea. According to Prof. Vijay's blog, one of the feature of v7 is "the ability to use the FAH client to manage multiple machines easily". How that is done, is currently unknown so you will have to wait until the open public beta is available.

2) Something like a scheduler, right? If so, I believe that it has been requested so we will have to wait and see.

3) Unfortunately, this idea isn't at all feasible because due to the variety of hardware available, every system performs somewhat different than the other. Once you include overclocked CPU, RAM, GPU, the amount of data that needs to be collected, analyzed, processed to make those WUs available isn't really worth it at all. Currently, your best option is the -oneunit flag which tells the Client to finish the current WU and exit once it uploads the wuresult.

4) An interesting idea.

5) There is already a website (http://fah-web.stanford.edu/cgi-bin/mai ... pe=osstats) that gives the performance of CPUs, GPUs, PS3, etc. For individual systems, I think that you can look-up the original specs from the manufacture and calculate it yourself.

6) You can use a kill-a-watt meter and find it out yourself Feel free to report it in the FAH Hardware Forum.

Are you using a third-party monitoring program like HFM.net or FCI? Both have useful features and can create a web page with detailed info regarding your client status. HFM is great for instant information and FCI is great for trending over longer periods as well. I've run both in parallel for well over a year. Also, HFM has a good EUE pause email notification function. Both programs are being actively developed by the authors and both authors are keen and willing to incorporate good ideas into their work.

Yes, I just started using HFM.NET and thanks out to the author/donor for a great program! It would be flame-bait to go further on the subject.

The "select a WU size" can't work under the current model. A large component of the project is points-driven, and if there is any real or perceived disparity in points between files of different sizes, some projects won't have many clients on them. Whether a pure-total-points system is the best engine for the project is another topic for another day. It's OK for me to have fantasies though.
But this is desirable. It would allow donors to decide what amount of donation they want to provide. If I want to contribute 8 SMP jobs a night, I would have the option.

Think of it this way. If I were shopping for Spreadsheet software, I'd have a desired feature list. That's what this is.

Qinsp wrote:Here is my fantasy client, this is not a complaint:

1) All FAH data storage for all machines located on one machine. Z:/miscdata/fah/machine001/SMP/, Z:/miscdata/fah/machine006/GPU, etc.
2) Allows on/off timer of all machines from one application.
3) Allows WU duration cap: "Machine 003 - No WU's over 12hrs." "Machine 008 - No WU's over 24hrs."
4) Sends email out based on switches - "Alert - Machine 005 will not complete WU before requested time (08:00)", etc.
5) Reports estimated FAH gigaflops for each machine and total. (Fun is nice too)
6) Reports estimated power usage per machine and total.

1. You can store your fah data files where you want. The systray client supports a "Start In" location. Doesn't have to be the local C drive.

2. Windows has a task scheduler. Use that instead. You can even copy the .job file from one machine to the next.

3. Fah unit size is by work unit, not by work unit duration. If the machines are not folding full time, do not use a full time client like SMP. Use the CPU client instead. Folding part time is acceptable for CPU clients. Yes, smaller WUs might be more optimal for you, but breaking up WUs in to smaller chunks is not helpful to the project. Even if the WUs could be broken up, the extra load on the servers would slow down the results anyway.

4. Nice feature, but not really needed when you setup the correct client for your folding environment. 3rd Party tools can do this also. Not necessary in the fah client.

5. Fun is nice, yes, but science has always taken a higher priority, so such features will always get a lower priority. 3rd party tools can probably estimate your system flops anyway.

6. No other program on your computer can do that. How do you expect fah to do it? Buy a Kill-a-watt, or a fancy UPS system. Even that dedicated hardware can't do this for a group of computers, but at least you can measure one or more systems so you can see the power usage.

I'm not trying to rain on your parade of ideas (several were good, IMO), but the fah client tends to follow a minimalist design approach. Simpler is always better. And Science features come before bells and whistles. Yes, obviously they try to make it user friendly, but the more time they spend on client development is less time they have to work on the science to cure diseases.

Don't worry, I have an umbrella.

So you are saying the existing clients can't be improved because it would hurt science, private donors are better at coding, and stopping jobs in the middle is best. That's your opinion. I don't agree, and will leave it at that.

I have no problems expressing myself. I don't need your help incorrectly rephrasing and twisting the meaning of what I said.

I did say the existing v6.x client can not be improved, but only because it's feature locked pending the release of a v7 client, not because improvements would hurt science.

I did not say private coders were better or worse. They have different priorities. Science is first with fah, nice to have features come 2nd, and don't always make it in to the client. And so private coders sometimes add them. The scientists at fah are not professional coders, so the minimalist approach helps on several levels. Easier to code, easier to test, easier to update and maintain, and easier for the end user to install/configure. And yes, in affect, spending a lot of time programming fluffy features does take time away from the science of the project. That doesn't necessarily hurt the project, but at the least slows it down. That said, they do try to keep a balance between productive features, and nice to have features. The result is that many suggested features not in the v6 client are rumored to be added in the pending v7 client. However, there is little gain in duplicating features in the fah client that can be done by other software, like the Windows Task scheduler.

And I did not say stopping jobs in the middle is best. But it is a good 2nd place option when not running a PC 24/7, but only on CPU work units that have longer deadlines. All contributions to the project are welcomed.

Qinsp wrote:Don't worry, I have an umbrella.

So you are saying the existing clients can't be improved because it would hurt science, private donors are better at coding, and stopping jobs in the middle is best. That's your opinion. I don't agree, and will leave it at that.

Anybody knows that clients can be improved, and items from wish-lists like yours may find their way into a client, but certainly not quickly, and perhaps not at all.

I think it's fair to ask a question that the Pande Group will probably ask themselves when they review your list. Suppose that feature A costs X programming hours at Y Dollars per hour. Would it be better to spend X*Y dollars on feature A or spend X*Y dollars on Alzheimer's research? Unfortunately that question disqualifies a lot of very good suggestions. What's left is a minimalist client that does the basic things pretty well with none of what 7im called "bells and whistles"

If Stanford were selling business software, it would probably come with all those features and a license would cost several thousand dollars. Instead, we wait for some volunteer to come along and say Here are step-by-step instructions to how I set up Windows Scheduler to run FAH (or some other item on your list).