I've had several tasks over the last 24 hours which are taking more than twice as long to complete (25,000+ vice 10,000) yet are providing very poor granted credit. (152 claimed / 51 granted, 197.16 claimed / 16.75 granted, 244.54 claimed / 80.00 granted ).

All of the tasks are reported as being completed sucessfully but I'm wondering if I should be looking at my overclock.

System is a Q6600 @ 3750 with 4G memory running Windows XP (fully updated).

Computer can be seen here.... https://boinc.bakerlab.org/rosetta/results.php?hostid=826115

Anybody else having these kinds of issues?

I've found this thread... https://boinc.bakerlab.org/rosetta/forum_thread.php?id=4375 and reported the long running tasks I've had. If you have a look they are all from the same group, so I'm just going to start aborting any of those tasks which show up.

I reported something similar not so long ago and was told that this was due to the research nature of the project. Although testing is conducted to catch issues such as rouge code in the ralph project, it would seem the remit of the Rosetta project also covers behaviour analysis. As such new research methods are tested within the rosetta work units to try and come up with better and faster ways of doing things. While some of them work, others don't.

This may be because the new code dosn't play well with your hardware and was never exposed to something like it during the ralph testing or it may be they just went down a wrong road. Either way by aborting it and not letting it have a chance your making it harder to conduct the research as they won't know if it worked or not, and for some of the work units it will work and something will be learened from it. For the ones that go long still something can still be learned, but not if you go round aborting them.

https://boinc.bakerlab.org/rosetta/forum_thread.php?id=4388&nowrap=true#56204

go there if you want to see extact what was said to me.

---

I think it is unfair that if one task takes my system four times as long to crunch as another task, to NOT give the longer task proportionately greater credit.

More problematic to me, I run off-line. That means when I do connect, I need to fetch enough work to keep my system busy until the next time I connect. Rosetta was good in that it allowed me to set a "standard task time" for each task. When new tasks were fetched, they were enough to keep my system busy.

But BOINC calculates a "duration correction factor" that it applies to the tasks. As long as all Rosetta tasks adhered to the "standard task time", BOINC downloaded what I thought was a proper amount of work. But now that some Rosetta tasks take longer, BOINC is estimating that *all* Rosetta tasks will take longer, with the result that it downloads FEWER tasks when I do connect. Given that most of those downloaded tasks are not the longer ones, my system runs out of work that much sooner (and I have to make "unscheduled" connections to fetch more tasks).

PLEASE, not only give more credit to longer tasks, but also give them a higher run-time estimate. That will keep the BOINC client from mis-calculating the workload.
.

Hi all.

I've been bit'n by the credit problems to, I don't have a problem with tasks

that use a lot of memory or take longer to run if needed.

(it would be nice to know beforehand) I do have a problem when a task runs for

11hrs + (instead of 6hr runtime) and only grants 16 credits that's just not right.

I hope these problems can be fixed.

pete.

---

Wish we could get a definitive answer to these issues from a mod/admin. So far it looks like the tasks run fine up to around 95%, but then bog down. My normal run time is 3 hours so anything over 4 hours I've been aborting and carrying on. (I've stopped arbitrarily aborting them before they start).

Is that the way to go?

Is that the way to go?

I posted some guidelines here. And I posted some information about tasks that "bog down" as you say here.

---

Rosetta Moderator: Mod.Sense

Is that the way to go?

I posted some guidelines here. And I posted some information about tasks that "bog down" as you say here.

Thanks for the response, but I'm still not sure (other than report them) what you folks would like us to do about them.....

Should we let them try to run all the way through (best guess would be +50% time to complete the final 5% of the run), or will it serve the same purpose for us to abort it once it gets into the .010% / 5 min phase?

I would review the guidelines posted at the first link about how long a model should take. Then confirm in the graphic what model I'm on. If you are no longer on model 1, then generally it will finish close to your preference. But lately there have been some models taking days rather then hours. Basically, if you are on the same model for more then a couple of hours, and still on model 1, abort it. It is taking longer then planned, and may never complete that model to report anything of better use back. I mean it may run until ended by the watchdog after 3x your runtime preference (9 hrs by default) and yet still not have finished a complete model, so if you abort after 4hrs, you're saving some time and not going to provide a scientifically useful result (other then to point out something needs to be changed to eliminate this long-running model).

---

Rosetta Moderator: Mod.Sense