Page 1 of 1

Did FAH ever consider to move to BOINC?

Posted: Thu Mar 19, 2020 5:57 am
by fangfufu
Obviously FAH predates BOINC, did Pande group ever consider moving abandon FAH and run everything on BOINC? What is the reason behind maintaining separate FAH codebase? I can think of many myself, but it would be great to hear from people who are more closely associated with the project. Bonus question, does Pande group still exist? Does Pande group still run their experiments on FAH?

Re: Did FAH ever consider to move to BOINC?

Posted: Thu Mar 19, 2020 6:47 am
by Joe_H
Yes, there are some much older topics here on that. Basically they found too many incompatibilities in the way they were doing things and the way BOINC was doing things for it to be a good match.

As for the actual Pande Lab, there may be some people still handling things there, others with closer contacts can comment on that. But the various grad students finished up their degrees and Dr. Pande is working for a private company. What is behind F@h at this point is referred to as the F@h Consortium - https://foldingathome.org/about/the-fol ... onsortium/. A number of the researchers at these labs are former grad students or post-docs at Pande Lab. Dr. Bowman at WUSTL is the Director.

Re: Did FAH ever consider to move to BOINC?

Posted: Tue Jun 30, 2020 2:34 am
by bruce
BOINC was originally designed based on the needs of seti@home. (Yes, iI has come a long way since then.) If you intend to analyze the radio signals within a specific part of the spectrum from a small patch of sky. It makes no difference which analysis is completed first or last as long as they all get analyzed. Each one has a rather loose deadline, of course but finishing your assignment early earns zero bonus. FAH, on the other hand cares very much how quickly you finish your assignment and which assignments must be completed first. Those facts lead FAH to grant generous bonuses for rapid returns and penaleies for exceeding the deadline.

That difference in philosophies has led to a rather extreme differences in goals an methodologies that are essentially incompatible.

Re: Did FAH ever consider to move to BOINC?

Posted: Tue Jun 30, 2020 11:36 am
by NBR
bruce wrote:BOINC was originally designed based on the needs of seti@home. (Yes, iI has come a long way since then.) If you intend to analyze the radio signals within a specific part of the spectrum from a small patch of sky. It makes no difference which analysis is completed first or last as long as they all get analyzed. Each one has a rather loose deadline, of course but finishing your assignment early earns zero bonus. FAH, on the other hand cares very much how quickly you finish your assignment and which assignments must be completed first. Those facts lead FAH to grant generous bonuses for rapid returns and penaleies for exceeding the deadline.

That difference in philosophies has led to a rather extreme differences in goals an methodologies that are essentially incompatible.
Great explanation. :)

Re: Did FAH ever consider to move to BOINC?

Posted: Tue Jun 30, 2020 1:41 pm
by Endgame124
bruce wrote:BOINC was originally designed based on the needs of seti@home. (Yes, iI has come a long way since then.) If you intend to analyze the radio signals within a specific part of the spectrum from a small patch of sky. It makes no difference which analysis is completed first or last as long as they all get analyzed. Each one has a rather loose deadline, of course but finishing your assignment early earns zero bonus. FAH, on the other hand cares very much how quickly you finish your assignment and which assignments must be completed first. Those facts lead FAH to grant generous bonuses for rapid returns and penaleies for exceeding the deadline.

That difference in philosophies has led to a rather extreme differences in goals an methodologies that are essentially incompatible.
Boinc can be used for protein folding with quick turn around times - check Rosetta@Home - so it’s not just focused on the seti use case. That said, I’m pretty sure Rosetta is fairly different than F@H in end folding goal as well.

Re: Did FAH ever consider to move to BOINC?

Posted: Tue Jun 30, 2020 3:11 pm
by Jonazz
Rosetta does not/barely simulate the protein folding process. They predict the 3D structure of proteins computationally and design new proteins.

Re: Did FAH ever consider to move to BOINC?

Posted: Mon Jul 27, 2020 1:30 pm
by toTOW
I think GPUGRID has a similar behaviour as FAH (next WU need previous one to be generated) ... unfortunately, it has lead to an elitist project only working well on faster and newer NV GPUs ... :(

Re: Did FAH ever consider to move to BOINC?

Posted: Mon Jul 27, 2020 4:42 pm
by JimF
toTOW wrote:... unfortunately, it has lead to an elitist project only working well on faster and newer NV GPUs ... :(
That is fortunate, if you are trying to get the science done. I just hope that FAH get a CUDA app soon.

Re: Did FAH ever consider to move to BOINC?

Posted: Mon Jul 27, 2020 8:15 pm
by MeeLee
bruce wrote:seti@home... It makes no difference which analysis is completed first or last as long as they all get analyzed. Each one has a rather loose deadline, of course but finishing your assignment early earns zero bonus. FAH, on the other hand cares very much how quickly you finish your assignment and which assignments must be completed first. Those facts lead FAH to grant generous bonuses for rapid returns and penaleies for exceeding the deadline.
..
It's not entirely true. If you return a WU early, it means you can start a new WU early. Which means more work can be done, which usually leads to higher points anyway.

QRB brings up the possibility of 2 users (one with a fast system regularly pausing his computer, the other with a slow system continuously folding non stop) doing the same amount of work, over the same amount of time, but one gets more points than the other; just for owning faster hardware; even if the same amount of work is done in the same amount of time.

Same is true for 2xRTX 2060 vs 1x RTX 2080Ti. They're both as fast as one another, get the same amount of data processed per time, but the 2080Ti will just get more PPD for it.

Re: Did FAH ever consider to move to BOINC?

Posted: Thu Jul 30, 2020 5:09 am
by bruce
MeeLee wrote:Same is true for 2xRTX 2060 vs 1x RTX 2080Ti. They're both as fast as one another, get the same amount of data processed per time, but the 2080Ti will just get more PPD for it.
You get more points because each WU is completed faster.

E.g. Run one project from t1 to t2 and another project from t3 to t4. You have removed one project from the server for t2-t1 and another for t4-t3. during which times nobody else can work on them.

Compare that to running two projects from T1 to T4. You have remove two projects from the server for longer periods of time so you've created more interference with other people ability to work on the server's pool of FAH's WUs. In the grand scheme of things, the latter means FAH needs to create more clones to distribute to donors even if the science doesn't need them.