Discussion: what is holding F@h back?

[[WHGT]Cyberman]

I'm trying to keep it short. I agree with almost all in this thread. Repetition for emphasis:

V7 client is an improvement, but it can go further.

The client must be able to help with problems.
Example:
GPU outdated, not supported anymore. Client doesn't know it, just reports errors.
User is confused, must look through half the internet to find out what happened and why.
The client should have said: "Sorry, your GPU is outdated, see this blog post for details."
I didn't know about the blog. I knew about the forum, and had to search here for a while to finally find this thread explaining what happened and why the client couldn't help.

F@H Output: More details, more information. Make them more accessible, explain their value better. (100 Papers doesn't look like much to ignorants like me. Is there more?)

Project pages need more information. "Markov State Model of Amyloid Beta" - nice. But what does it mean? Link to Wikipedia for explanations, perhaps? More details about what it's for, better description of what it does.

Select projects to fold for - not just one, multiple. People WILL cherry-pick, better to give them more than one choice. Have them select "primary project" and "secondary" or something.
Display that information on group pages as well! Give them something to rally behind.

Make it easier to keep track of what you folded. Perhaps integrate/bundle 3rd party software? If possible, overall progress of a project, especially if you're involved in it. (Finished runs perhaps?)

Mobile devices. Android is almost everywhere. These devices cannot do much in itself (yet) but they are legion. I'm constantly carrying a phone and an MP3 player - both android. They could do work when charging.

--

I tried to keep it short, didn't want it to sound harsh, though. I still fold, I will continue to do so. I just think there is lots that could/should be done.

[k1wi]

Technically you can change the number of CPU cores at any time the client is paused. However, it is recommended to finish a WU before changing the client's number of CPU cores.

I use a very basic bash script that pauses/unpauses slots based on the wholesale price of power here using a range of Linux tools including expect, awk and a few other goodies. The script runs on a Linux server that amongst other things such as the price of power, logs the temperatures of the Windows computer (it reads the values coretemp logs to file). Because the wholesale price of power is only made publicly available every five minutes I used this script as part of a wider cron job that runs every five minutes.

I modified it to set it to pause when the computer hit a certain temperature and it can be found here: viewtopic.php?p=230919#p230919

[Jesse_V]

Those are some excellent points [WHGT]Cyberman!

Project pages need more information. "Markov State Model of Amyloid Beta" - nice. But what does it mean? Link to Wikipedia for explanations, perhaps? More details about what it's for, better description of what it does.

There's a lot of variation in the project descriptions. I wish they were more consistent. Realize that there's a lot of scientists using F@h and some are better than others at converting everything to laymen terms. Just so you know, here are some explanations of Markov State Models:
http://en.wikipedia.org/wiki/F@h#Project_significance
http://folding.stanford.edu/English/FAQ-Simulation

Select projects to fold for - not just one, multiple. People WILL cherry-pick, better to give them more than one choice. Have them select "primary project" and "secondary" or something.
Display that information on group pages as well! Give them something to rally behind.

There's already a Cause Preference in the client. Under the Configure window, in the Advanced tab. Doesn't seem to affect WU assignments at this time, but it's very presence indicates that it should work "soon".

Mobile devices. Android is almost everywhere. These devices cannot do much in itself (yet) but they are legion. I'm constantly carrying a phone and an MP3 player - both android. They could do work when charging.

Heat could be a problem. I have a GPU-accelerated Mandelbrot Fractal app on my Android phone, and the phone itself gets pretty hot when I'm playing around with the app. There's no cooling fans; it's just passive heat dissipation.

[csvanefalk]

As legion as Android devices are, it is dubious they could do any useful work. If you let them fold 24/7, they WILL drain your battery (and possibly overheat if they are in you nice cotton pocket) in a matter of hours. If you just let them fold while charging, they will most likey need very small WUs to meet deadlines. In either case (and especially the latter), also remember that each TCP connection to FaH servers costs as well. Androids could rack up a whole lot of them.

[bruce]

When WU is in progress and you change number of cores (-smp #), WU will continue from 0% (from the beginning). I have tried this a long time ago, and now I guess so.

I believe that has been fixed. It seems to resume work correctly for me now though changing the number of cores in the middle of a WU is not recommended.

[[WHGT]Cyberman]

There's a lot of variation in the project descriptions. I wish they were more consistent. Realize that there's a lot of scientists using F@h and some are better than others at converting everything to laymen terms.

Well, all I see is project 11293 and recently 11292
I did look the terms up on Wiki, but I don't think everyone is going to do so (if people don't even look up what's in their food - E-Numbers I mean - why would they look up what their computer is doing? Unless it's just a click away.).

There's already a Cause Preference in the client. Under the Configure window, in the Advanced tab.

Not in my client. Perhaps too old - 7.1.52

As legion as Android devices are, it is dubious they could do any useful work.

They are going to improve - android likely isn't going to be limited to cell phones and small mobiles. Linux is almost everywhere already, most might be replaced with android, offering a more or less homogenous interface.
But yes, for now at least they'l likely only do little work. Perhaps some of the recombination workload could be distributed as well? Have special clients that take in partial WUs and combine them before sending them off. No single client could replace a server, but perhaps they could lighten their burden. How tight is internet bandwidth nowadays? Does the majority still pay by volume, or flat rate?

[csvanefalk]

They are going to improve - android likely isn't going to be limited to cell phones and small mobiles. Linux is almost everywhere already, most might be replaced with android, offering a more or less homogenous interface.
But yes, for now at least they'l likely only do little work. Perhaps some of the recombination workload could be distributed as well? Have special clients that take in partial WUs and combine them before sending them off. No single client could replace a server, but perhaps they could lighten their burden. How tight is internet bandwidth nowadays? Does the majority still pay by volume, or flat rate?

Android is developed specifically for tablets and mobile devices, so I doubt you will see it having any significant spread beyond that, especially since it contains proprietary components. That said, since Linux 3.3 the two kernels have begun to merge into a single kernel (again), so a homogeneous interface is more likely to come from growing shared library support.

Unfortunately there is no way to divide WU:s, the work is strictly serial (i.e. the calculation of one frame depends on the frames preceding it). One could wish it was possible since it would also open up for folding on clusters, which would be very cool.

Internet bandwidth is not really the issue, since run-of-the-mill WUs (and their results) are not really big by contemporary internet standards. I would say the vast majority, at least in the west is on strict flatrate. A bigger problem is the amount of pressure put on servers with multiple incoming connections. Too many small WU:s will result in a lot of those.

[k1wi]

They are going to improve - android likely isn't going to be limited to cell phones and small mobiles. Linux is almost everywhere already, most might be replaced with android, offering a more or less homogenous interface.
But yes, for now at least they'l likely only do little work. Perhaps some of the recombination workload could be distributed as well? Have special clients that take in partial WUs and combine them before sending them off. No single client could replace a server, but perhaps they could lighten their burden. How tight is internet bandwidth nowadays? Does the majority still pay by volume, or flat rate?

Android is developed specifically for tablets and mobile devices, so I doubt you will see it having any significant spread beyond that, especially since it contains proprietary components. That said, since Linux 3.3 the two kernels have begun to merge into a single kernel (again), so a homogeneous interface is more likely to come from growing shared library support.

Unfortunately there is no way to divide WU:s, the work is strictly serial (i.e. the calculation of one frame depends on the frames preceding it). One could wish it was possible since it would also open up for folding on clusters, which would be very cool.

Internet bandwidth is not really the issue, since run-of-the-mill WUs (and their results) are not really big by contemporary internet standards. I would say the vast majority, at least in the west is on strict flatrate. A bigger problem is the amount of pressure put on servers with multiple incoming connections. Too many small WU:s will result in a lot of those.

The underlying architecture (GROMACS) allows clusters - for example you can compile it on Blue Gene. But seeing as most users don't have access to a super computer they [FAH] have chosen to avoid clustering

I can't see ARM based Android/iOS running F@H on the CPU. One day maybe on the GPU if it becomes powerful enough & standardized enough.

For a really good discussion on mobile DC compute it have a look at this thread: viewtopic.php?f=20&t=22241. 10 hours for a Milkyway WU on a phone plugged in or 1 minute on a GPU?

[JimF]

On a more prosaic level, all I ask is the ability to buy an additional graphics card with some assurance that it will work. I long ago gave up trying to Fold on my display card, though the screensaver idea is a good one and it is a little puzzling to me (a non-programmer) that it has not been implemented.

But I have just spent another morning trying to get an HD 7770 (non-display) card to fold along with a GT 240 display card that won't be folding. It didn't work; the identity of the cards kept getting mixed up, etc. I have too many other stories like that for this thread to be bothered with, some with Nvidia cards, others with mixed Nvidia/AMD cards. The only thing that works reliably for me is to use either a single card, or else two matched cards. But as noted above the single card does not work on my main PC where I need a dedicated display card, and using two cards makes upgrading difficult since you normally don't want to buy two identical cards at once.

As someone else noted in another discussion, BOINC has no trouble using multiple cards with different GPU types, and so I have reluctantly turned the HD 7770 over to BOINC duties. But I really would like to buy a Kepler (GTX 650 Ti) as a non-display Folding card while using either of the previous cards as the display card, but I have no assurance that it will work. I have no problem spending money to keep up with new technology, but gambling is not my thing. I think this limits PG's audience more than it needs to be.

[vip00]

I have also turned to BOINC recently after a long time of working exclusively with F@H, largely because of an absence of a "fold while idle" option. I upgraded from a HD 4870 to a 660 GTX after F@H stopped handing out work units for the older architecture... However, now I get frame lag even in things like Flash videos. It's really unusable to any real extent with the GPU client folding. In the absence of an option to fold while idle or some version of GPU priority, I don't think I can reasonably keep contributing to F@H.

That definitely puts bringing back the screensaver as top priority for making F@H more usable!

[art_l_j_PlanetAMD64]

I have also turned to BOINC recently after a long time of working exclusively with F@H, largely because of an absence of a "fold while idle" option. I upgraded from a HD 4870 to a 660 GTX after F@H stopped handing out work units for the older architecture... However, now I get frame lag even in things like Flash videos. It's really unusable to any real extent with the GPU client folding. In the absence of an option to fold while idle or some version of GPU priority, I don't think I can reasonably keep contributing to F@H.

That definitely puts bringing back the screensaver as top priority for making F@H more usable!

Why don't you just use FAHControl to pause all folding when you are using the computer, and un-pause when you have finished using the computer?

It works for me.

[proteneer]

Interesting read on your opinions. Some of them echo our own concerns. Sorry that I can't elaborate at this point, but I appreciate the comments

[proteneer]

Ps. Our new core will hopefully alleviate some of these issues. We are quite keen on problem solving

[ChristianVirtual]

I understand that resources for the development of the software infrastructure are limited and potentially even restricted due to the source of funding. Skilled developer also not growing on trees (though I would think Stanford has a bigger pool then others).

Of course I accept that the primary target for most developments is to get science done. That include core development for CPU & GPU and the projects/WU we are chewing day in and out.

But reading on the front page we want to reach 1'000'000 donors I wonder what we can do to get the missing 700'000; beside talking with friends and communities.

I remember my first days trying to understand the setup of FAHControl and connect to other folding boxes. Could have been more intuitive for newbies like me at that time. That was make me wonder what could be done to make it more end-user friendly. Not talking about the technical guys; talking about Joe Doe.
Would investments in more state-of-the-art front end tools/infrastructure attract more Joe Does and eventually get more science done by reaching 1'000'000 folder ?

Not talking here about concrete ideas; first I would like to understand if there is a process for change management in place, how it works, who is involved and how to get involved properly. And what are the overall primary and secondary goals.

[7im]

First step is to join the beta team.

Then review: viewtopic.php?f=86&t=21367