Just joined this project. Upon downloading a number of work units I noticed that two of them would take about 40 + hours to completion but most were in the 5 hour and 30 minute time frame. Is this normal?

---

Rosetta has "configurable runtime WUs" where Rosetta keeps creating as many new predicted 3D protein structures, as will fit in the time specified.

As the minimum work is 1 Model (structure), some rare large WUs might take more than the default (especially on slower PCs).

You can learn more in the FAQ on WU runtime

AFAIK, for the last month default WU runtime has been 2hr/WU. Personally I run 8hr WUs and prefer to keep a very small WU cache of just 1-3 WUs per P4 PC, by setting BOINC time to connect to 0.1 day.

---

Best UFO Resources
Wikipedia R@h
How-To: Join Distributed Computing projects that benefit humanity

Thanks for the info. I understand the system now.

---

...
Personally I run 8hr WUs and prefer to keep a very small WU cache of just 1-3 WUs per P4 PC, by setting BOINC time to connect to 0.1 day.

Cool! That means you get the new bugs first.

---

dag
--Finding aliens is cool, but understanding the structure of proteins is useful.

Well, in my 3.5 months of crunching with 3 * P4s (XP and Linux), a couple of days ago I just encountered my 2nd bad WU.

That's why I tend to think many of the problems could be hardware-related, or due to misunderstandings (as soon as a WU takes a bit longer than 2hr like *largescale*, they are quick to abort it).

As long as one has 24/7 Internet and PC operation, I don't see why not set BOINC to connect every couple of hours (0.1 day), like I did, as it'll reduce WU round-trip-time and help more with "smart" sampling.

Copying my comment in Science forum:

3/ Ideal WU round-trip times, considering "smart resampling strategy". I mean, if one "loads the boat" downloading 60 * 2hr (default WU runtime) FARELAX_NOFILTERS_xxxx WUs to "last" him for 3-4 days, EVEN IF HIS PC REPORTS BACK EVERY 2hr after each WU is finished, many won't be useful because after the first 10k results we may focused on a different part of conformational space, right? And the least desirable "behaviour" would be reporting all 60 WU together after 4 days (because that PC will probably have spent its time looking randomly, rather than being "smart" and focusing on the most "promising" space).

1. Testing a smart resampling strategy:

FARELAX_NOFILTERS_xxxx
FACONTACTS_RECENTER_NOFILTERS_xxxx
Have you ever wondered whether there's a better way to fold proteins than have 100,000 clients do completely independent runs, without talking to each other?
These workunits are testing a new strategy that may be smarter. We look at the full-atom scores and conformations from the first 10,000 models (that's the first workunit). For generating the second round of 10,000 models we then adjust the initial stages of the search (which uses a low resolution score function) to favor contacts that gave low energies in the first round.

If I got it right, I think this is another thing to be handled by BOINC server scheduler/feeder, so it sends WUs which necessitate a fast round-trip-time only to PCs which work 24/7 with permanent Internet connection and report back every few hours.

---

Best UFO Resources
Wikipedia R@h
How-To: Join Distributed Computing projects that benefit humanity

Since you are a new Rosetta cruncher I was thinking it was due to your Rosetta benchmarks not being properly established yet.

---

Founder of BOINC GROUP - Objectivists - Philosophically minded rational data crunchers.