IBM Grid Near 50,000 machines - Slashdot Users #13

another similar writes "IBM's World Community Grid is off to a roaring start. Since kicking off six weeks ago (original Slashdot story), the grid has grown to almost 36,000 users with almost 50,000 machines. Growth continues as more media coverage hits. There is a team of Slashdot users - currently ranked 13th in points with only 79 members. If you have spare cycles, download the software, join us and crank for medicine. For those of you with dual processor systems, you'll have to use a homebrewed tool - beyond two is not supported yet. Alas, you also have to be running Redmond's finest. According to their FAQ, a Linux client is slated for development in 2005."

Comments

Ownership FYI

[The Hobo]

For those keeping track,

"IT IS ANTICIPATED EVENTUALLY THAT IBM WILL TRANSFER WORLD COMMUNITY GRID TO A NON-IBM OWNED WORLD COMMUNITY GRID ENTITY OR ANOTHER ENTITY"

This is from the license. Just something to watch for in the future if you like Big Blue (or don't like them) or are concerned about it.

Rosetta was developed on Linux

[goombah99]

I'm one of the authors of the code they are running as the first application of the world grid. This is Rosetta, the protein structure prediction program. Rosetta was born on Linux. It can run on a mac too but not as well. There never was a version developed for Windows. But hand it to the the IBM folks to create a wrapper that lets it run as a grid "screen saver" scavenger application on windows. Pretty remarkable.

Of course the reason for this is obvious right? windows dominated the planet not only in installed systems but in installed systems with cycles to spare. i.e. desktops. So dont cry your eyes out over it not being linux compatible. The excess linux bandwidth after you subtract our the servers is not going to be a lot. Console yourself that the TCO of linux is really a lot less when you figure that linux computers are already too busy to be bothered with Grid computing. :-)

Rosetta itself was written in fortran and only recently converted to C++. the C++ conversion was done using the incredibly well designed Objexx Library by stuart metzner and colleagues. This is a library that lets you write fortran code in C++. Before this people who tried to re-write this behemoth to C++ just died in the process. The objexx library let the whole thing be converted to C++ in one fell swoop. Now the program will slowly evolve from fortran style to C++ object orientation as it continues to grow. But in the meantime the code is productive. Nice Eh? The cool thing is that with a bit of optimization the code did not lose any appreciable speed in the conversion. So if you have legacy fortran you use for speed, consider converting it using Objexx. I was one of the people who argued for going to fortran95 not c++ because I feeared a speed loss; Iv'e become a convert

In any event the program is not like folding at home. That program tries to study in detail the picosecond evolution of single protien as it folds. Rosetta simply predicts the folded structure. Its actually quite fast at doing that. But it turns out it makes lots of different predictions. So you have to do it tens of thousands of times and then see which geometries of folded structures are favored statistically. Then you do the next protein. Eventually you work your way through the whole human genome.

also unlike folding at home the potential surface in rosetta is less physics based and more bayesian statistice. It has statistical potential for the probability of a peptide backbone structure occuring. And it has a probabilty for a sidechain amino acid sequence given a backbone structure. Multiply those together and bayes rule says the result is proportional to the probablity of a structrure given a sequence. You can read more about this here. Click on publications.

This statistical potential turns out to be so accurate that it can not only be used to predict the structure of proteins but it can be used in reverse to design a novel structured protein. Recently it was used to design a protein with a tolopology that had never previously existied in nature. This is rather an amazing results. Others had previously redesigned the sequences of existing topologies or perturbed those topologies or created some special case topologies. But Brian Kuklman in David Baker's lab actually started from a napkin sketch and designed a protein from scratch.

After you predict the structure of a protein, one thing you can do is ask if that structure is like another Protein you have seen before. You can compare the structure of a model to a real protein using a program known as MAMMOTH. While there are a variety of programs for comparing two proteins this one is particularly good for the case of comparing an inaccurate model to an experimentally known structure. If they match then you can assume the protiens may share a related function or evolutionary origin (or not!).

whihc brings us to what proteins are. Think of DNA as a disk drive that

Re:Only 79 /.ers in six weeks. What does that say?

[IO ERROR]

OK, you can mod me -1, Wrong and Stupid now. I pulled the last week of server logs to see what I'd come up with from people referred from /. and this is what I find. I truly deserve the flamefest which will follow...

65% Windows
23% Linux
10% Mac OS X
2% Everything else

User agents:

68% Firefox
11% MSIE
7% Safari
3% Opera

Firefox versions:

84% 1.0
10% 0.9.3 or older
6% 0.10.1