The Science of Folding@home

mr_sifter writes "As previously discussed, computers running Folding@home now contribute over 1 petaflop of processing power to research into protein folding, making Folding@home the most successful example yet of a distributed computing app. It's also at the forefront of GPGPU computing, with both Nvidia and ATI keen to push how well their graphics chips perform when folding. So the technology is great, but what about the science? This feature looks at how the Folding project was developed, how it's helping researchers and the thorny question of how long it might be until the software running on your PC or PS3 actually produces real-world results."

Comments

i use folding@home

[ionix5891]

well its more like folding@office and making better use of the taxpayers money (research facility workstations)

but one thing bugs me

has anyone done the maths as to the electricity used by folding@home so far? the servers i run this on when i go home are always at 100% and by time i return in morning the office is nice and warm, since im not the one paying for the electric i dont really care

im not really sure this project is "green" is what im trying to say

Re:i use folding@home

[Futile Rhetoric]

Let me Wikipedia that for you.

Re:i use folding@home

[fuzzyfuzzyfungus]

The (somewhat trickier) question, is not "How much energy does folding@home use?"; but "How does folding@home compare to other methods of doing the same calculations?".

As long as we accept that doing the folding is a worthwhile use of resources(which, unless we are busy communing with the moon goddess or wearing uncured leather and killing bears with our teeth, is probably agreeable to most) the question is a matter of how to do it most efficiently; balanced by the fact that sometimes doing it inefficiently is the only way to do it.

Unfortunately, I suspect that folding@home might fall into that category. If everybody participating were able to total up the costs they incur by doing so, and just donate that to the project, you could probably get better results by buying hardware well matched to the task. Unfortunately, because of transaction costs and psychological factors, and people who don't (directly) pay for electricity, it is much easier to get "in kind" donations of CPU time, even if they are less efficient. It's rather like bittorrent that way. Looking at the costs across the network, it'd almost certainly be cheaper to have Akamai or Amazon host the stuff, and have downloaders pay $.50 or so, rather than keeping their computers on for hours in order to pay in their (limited) upstream bandwidth. However, donations in upstream bandwidth are quite easy to collect, while handling money introduces complexity.

Re:i use folding@home

[jowilkin]

I think it's debatable if using research facility workstations for FAH is a good use of taxpayer money. What about powering them down instead?

I have been wondering about power consumption of these distributed computing projects for a while. How do you justify the sheer amount of energy used to run these things?

SETI is a much more questionable use of power IMO, but Folding@home has not really shown to be enormously useful considering the amount of power it uses.

Why not put all the money used powering computers involved in FAH into innovative research grants instead? Granted this is logistically much harder than convincing people to install a program on their computer, but it would be much more effective in furthuring cancer research.

Re:i use folding@home

[Anonymous Coward]

If you leave the server on anyways, the power draw at 0% usage versus 100% usage is minimal.

eg. I have an old Dual processor Dell server that uses 170 watts idle, and 180 watts running at 100%

the largest difference I got was on my overclocked AMD FX-55 which had a difference of 30 watts. and a laptop that had a difference in 30 watts.

I used the Kill-A-Watt device purchased from thinkgeek.com.

Re:i use folding@home

[JustinOpinion]

If everybody participating were able to total up the costs they incur by doing so, and just donate that to the project, you could probably get better results by buying hardware well matched to the task.

Maybe or maybe not. One would have to include in the calculation the cost of building additional computers. One of the ways in which distributed computing is "green" is that it uses computers which have already been built, but would otherwise be idle. In this sense it is re-using resources that have already been committed, rather than requiring totally new equipment to be built, which consumes new resources.

In other words, the newly-built computers would have to be sufficiently more efficient that they fully offset their own production costs, and then some.

As long as we accept that doing the folding is a worthwhile use of resources ... the question is a matter of how to do it most efficiently

Indeed. For many computational projects one has to take into the account the likely scaling of computer power and algorithmic power. For instance in principle for a given problem with a given deadline, it can sometimes be cheaper to wait until new computers are on the market, if they will be sufficiently faster (at a given cost) than the older ones. (That is, you may be able to "waste time" and still make your deadline.) Alternately, it may be a total waste of modern computer resources to inefficiently search a given parameter space if we have reason to believe that drastically better algorithms will become available in a few years.

As it turns out, problems like protein folding are very difficult, and we have no reason to believe that dramatically better techniques are on the horizon. So if (as you say) we care about the problem at all, then it would seem that we can justify the energy spent doing those calculations right now on modern general-purpose machines.

Re:i use folding@home

[OeLeWaPpErKe]

Heh, if you're truly worried about that, worry about treatments. If any (non-trivial) treatment consisting of specifically folding proteins is found, then there will be exactly one way to produce said drug : genetic manipulation. Only a genetically modified cell will be able to produce those custom proteins.

So it's not "buzzword-compliant" in more than one sence. It burns heaps of co2, it relies on genetic modification, specifically on injecting live humans with substances coming from "mutants" (just like most insulin today, but hey, at least that protein was natural in origin).

And it's also a sign of "skynet" to come, so to speak. Right now humans are telling these computers what to do, but there is no way in hell a human could ever hope to do what these supercomputers do. In 10 lifetimes you wouldn't get 1 nanosecond from the initial chain using paper.

This is basically a computer using a (basic) kind of artificial intelligence to respond to human questions about the real world. I doubt AI of this level would become self-conscious any time soon, but if an intelligence were to directly (or through deception) control a ribosome, like this one does, that would enable it to self-replicate. The question it would need to answer is one that is "but" an exercise in protein folding : "how do I fold a protein so it runs my thinking algorithm ?", even if it's much harder than the current questions being asked.

And if said AI wanted to create new weapons against humans, here's a quote from the article :

Pande explains that this is "as drug design is very hard, itâ(TM)s very easy to do more harm than good, and thatâ(TM)s one thing that we never want to see".

It's perhaps worth mentioning that the DNA code that defines AIDS, for example, is only a couple of thousand basepairs long, well within reach of this program.

Like most research worth pursuing, it's very, very not-buzzword-compliant, and conceivably unbelievably dangerous. Even something "undeniably good" like fusion research produces (tiny amounts of) long-lived high-radioactive waste due to a process called activation.

Re:i use folding@home

[xrobertcmx]

One of the things I did to resolve this at my home was to build a reasonably powerful 100W server. I use this machine to serve video to my PS3, as a Samba file server, and a few other things, but since I already wanted it on 24/7 I also run the smp package for folding at home on it. Total power used is 118W per the killowatt.

Re:i use folding@home

[Amouth]

it would be intresting but remember that cray is now selling GPGPU powered mini super computers.

Re:i use folding@home

[Thaelon]

Folding@Home, and torrents are more like a micropayment system that actually works.

Sure it costs you electricity and bandwidth, but in such small amounts (typically) and over time. Plus there are no additional transaction fees or middle men taking a cut. It's tax free, and there are no forms to fill out or any other bureaucracy.

Torrents are just a distributed micropayment system.

Re:i use folding@home

[compro01]

According to my UPS's monitoring software, by machine idles (screen off with torrents running) at 180W and hits 210W with folding@home (GPU edition on my 8800GTS) running, 30W extra. Assuming it does that 24 hours a day (it's always on anyway as it also runs as my FTP/web server and for remote access), that comes to 21.6KW-hr per month, which at local electrical rate (9.6 cents/KW-hr) comes to $2.07 per month.

Re:i use folding@home

[fast turtle]

Strange as I built my desktop two years ago with energy efficiency as the primary goal. I'm currently running a 64bit multi-lib version of Gentoo on a C2D e6300 (1.8GHz) with 8GB of memory and F@H using the 64bit SMP version set to Large (>10M) work units and my system is using a grand total of 120 watts average (that includes my LCD monitor and Linksys WiFi router) according to my APC battery backup. Hell I rarely turn my system off so it makes sense to run F@H and use the CPU while my system isn't being used for much else.

Based on my normal usage, I could actually get by with a 700Mhz Celeron and Win95 if it weren't for flash and Folding.

Re:i use folding@home

[ID000001]

Keep a few things in mind:

1) A good potion of the computing power for folding@home, are from PS3, which, according to study, are one of the most efficient CPU design along the top500.

2) Computer or Console that are running Folding@home are usually power on anyway. While increasing it's CPU utilization, the actual increased electricity use are likely lower then, say, a dedicated server just for folding.

3) It is very spread out, instead of everything running in a data center, everyone runs one at their home. Think of this as some kinda "heatsink" so less power are used to keep this system cool down, compare to, say, concentrated computing power in the same location.

4) It operate on many electricity grids. When it is ran across the global, it put less strain on forcing certain power plant to runs at higher capacity. I don't know if this make it more or less power efficient, but it would put less strain on individual power grid.

5) It give everyone a chance to contribute. They have a common enemies that killed many of their relative. They fight together, by themselve or in team. They keep people motivated on something. That count for something, right?

Not a petaflop!

[tomknight]

It's petaflops, not petaflop. That s means something.

Re:Not a petaflop!

[noundi]

Actually it's petaFLOPS as it's an abbreviation.

Re:Not a petaflop!

[Anonymous Coward]

Floating point operations per second; if anyone was wondering (honestly how long will you guys fight about that before actually saying the name?)

Re:Not a petaflop!

[master5o1]

No, like 1*10^15 "Floating Point Operating Per Second"

Re:Not a petaflop!

[master5o1]

*Operating==Operations
FLoating point Operations Per Second

Re:Not a petaflop!

[Useful Wheat]

Flops is short for FLoating point Operations Per Second. There is a point to the s.

5 petaflops, not 1

[Anonymous Coward]

"However, even Roadrunner looks decidedly weedy compared with the power of Stanfordâ(TM)s Folding@home project. Its computational power has now surpassed the five petaflops mark."

Re:5 petaflops, not 1

[wagnerrp]

The thing is that Roadrunner can run programs that are not ridiculously parallel. Folding is split into discrete units that are computed and sent back up to months later. The units have absolutely no bearing on anything else running, and there is no need for intercommunication. If you had to perform some task that could not reasonably fit within the free resources of an idle computer, you're sunk. Sure, it has a huge amount of capacity, but you cannot compare it to something like the Roadrunner because it cannot, and was never intended to, perform the same sorts of tasks.

Re:BOINC?

[jedirock]

It's more general purpose, and you can make your own UIs for it. I'm in the middle of doing one myself.

I use Rosetta @ home and foldit

[goombah99]

Rosetta at home is another and arguably much more efficient folding project. It actually predicts protein structures at high resolution, allows docking, and design of proteins. put your cycles there. Also if you like this kind of thing then try out foldit. it a multiplayer game in which you race others either collaboratively or in cometition to fold proteins. The games are chosen so the answers help investigators studying the protein folding process! The idea is to separate what humans do best--large scale long range geometry-- with what computers do best--fine tuning interactions.

Re:I use Rosetta @ home and foldit

[goombah99]

The main difference between folding at home and roestta@ home is that folding at home studies molecular dynamics-- the science of how proteins vibrate and move while rosetta actually goes after protein structure itself directly. As a result Rosetta can fold proteins with millions of times less computation.

It's actually useful for a PS3

[Dr. Manhattan]

Sometimes I need to leave my PS3 on for a while. (Recharging the controllers, big download, whatever.) I have the "automatically turn off after one hour of inactivity" setting checked, so sometimes the process wouldn't finish before it shut down.

So, I fire up Folding@Home (technically called "Life With Playstation" now) before I go to bed. Takes about six hours, plus or minus. Enough time for downloads or recharging, does something useful while the PS3's on, shuts off once the work unit's done, everybody's happy.

I prefer Seti@Home

[Drone69]

That way once we find the little ETs they can fold our laundry for us. Oh...hang on a minute. THAT Folding@home!

BOINC sucks, btw.

I preffer aqua (Adiabatic QUantum Algorithms)

[Seth Kriticos]

I donate my idle processing power to the aqua@home project ( http://aqua.dwavesys.com/ ). They (d-wave) are building quantum computers and that's a field I'm more familiar with than medicine. Guess both are more sensible than looking for E.T. though. (Just my personal opinion.)

What a Revelation!

[Prototerm]

And here I thought that folding@home had something to do with laundry. Who knew?

Paranoid much?

[Sporkinum]

My paranoid mind wonders how we really know these CPU cycles are working for good and not evil? It could be decrypting keys for all I know, or working out some sort of weapon system. We just have their assurances that a "work unit" really is going toward something worthwhile, and not to the CIA or NSA.