Harmonious Trees 0.03

[fwjmath]

Hi Beyond,

Your workunit will be finished in a few hours, probably 5-6 hours, just after it finishes all trees with first subtree 0123456745, and it position now is rather near the end.

fwjmath.

[Beyond]

Hi Beyond,

Your workunit will be finished in a few hours, probably 5-6 hours, just after it finishes all trees with first subtree 0123456745, and it position now is rather near the end.

fwjmath.

Thanks for the info. It completed in 94 hours.

[Dirk_P_Broer]

http://www.rechenkraft.net/yoyo/result. ... d=12426046 http://www.rechenkraft.net/yoyo/workuni ... d=10030836 WorkunitId 10030836] kept adding 3 seconds of time to completion for every second it ran, having a progress after 21 hours of 0,001% and wanting 76 hours more. It had no ckpt.txt file, so I killed it after 89,162.66 seconds of CPU time (already the second longest for this WU).

I've had another Harmonious Trees WU last month that kept on running when the estimated time to completion was zero. It had a progress of 0,000 and so I killed it after 29 hours. It had no ckpt.txt file either.

[Ananas]

Never kill _R_ workunits if they do not show progress, they might even run for 3 days (Yoyo's days seem to have 35 hours ) on a slightly OC'ed Q9400 with only 0.002% progress.

Those _R_ things are continued runs from previous workunits that had reached their maximum runtime and that usually happens in a phase with a very (v e r y) slow progress, which might even require a third (fourth? fifth?) workunit for the same calculation.

[Beyond]

Never kill _R_ workunits if they do not show progress, they might even run for 3 days (Yoyo's days seem to have 35 hours ) on a slightly OC'ed Q9400 with only 0.002% progress.

Those _R_ things are continued runs from previous workunits that had reached their maximum runtime and that usually happens in a phase with a very (v e r y) slow progress, which might even require a third (fourth? fifth?) workunit for the same calculation.

Thanks for this info! I was debating whether to abort one of these that has been at 0.015 progress for 30 hours. I'll let it run.

[fwjmath]

http://www.rechenkraft.net/yoyo/result. ... d=12426046 http://www.rechenkraft.net/yoyo/workuni ... d=10030836 WorkunitId 10030836] kept adding 3 seconds of time to completion for every second it ran, having a progress after 21 hours of 0,001% and wanting 76 hours more. It had no ckpt.txt file, so I killed it after 89,162.66 seconds of CPU time (already the second longest for this WU).

I've had another Harmonious Trees WU last month that kept on running when the estimated time to completion was zero. It had a progress of 0,000 and so I killed it after 29 hours. It had no ckpt.txt file either.

Hi Dirk_P_Broer,

It is somehow troubling that your workunit had no ckpt.txt. We can prove that, assuming everything outside the code (e.g. Windows library, compiler, etc.) works as expected, the application will always checkpoint every at most several minutes. Your ckpt.txt should be under some slot/n (n is a number). If it doesn't checkpoint, then I think something is going very wrong. You might want to check your machine.

fwjmath.

[Ananas]

... If it doesn't checkpoint, then I think something is going very wrong. You might want to check your machine. ...

I had one that didn't modify the (existing) checkpoint file for quite a while (had mentioned it somewhere else in this forum) but after a few hours, it decided that it would be a good idea to return to the standard checkpoint interval. That result has been a success and did not cause any problems for the validator either.

That looks to me like some loop / iteration in the calculation that takes way more time than expected, so it does not meet the checkpoint procedure while it is busy.


For an _R_ result, this might well happen at the beginning of a calculation, so the first "modification" (which would be the initial creation in this case) is delayed.


I don't think that this is necessarily an indicator for a problem, neither hardware- nor software-sided.


Imo, the only situation that could cause trouble would be, that this checkpoint delay lasts longer than the maximum runtime (~3 days).

[fwjmath]

... If it doesn't checkpoint, then I think something is going very wrong. You might want to check your machine. ...

That looks to me like some loop / iteration in the calculation that takes way more time than expected, so it does not meet the checkpoint procedure while it is busy.

Ah, thanks Ananas, you may be right. I have to check my code to find a way to make it checkpoint. Though it may not affect the result, but volunteers (and I) usually don't want to see things irresponsive, and prefer things to be checkpointed frequently and regularly.

In fact my code is originally designed for n=30, as computation time scales up exponentially (though with a mild base), it may introduce overly long running time between checkpoints in some rare case. I shall try to do something about this issue.

fwjmath.

[Nature Boy]

I've had some long running WUs but I now have one that appears that it is going to run longer than its deadline. The deadline is 10:05 pm this evening and it is now 3:30 pm, but it is not running high priority. It has been running a little over 37 hours and has 0% completed and no time to go. Should it be aborted, or should I let it continue?

I don't have a problem with long running WUs, but if this one is just wasting time, would make sense to stop it and allow other Harmonious Trees WUs to run.

[fwjmath]

I've had some long running WUs but I now have one that appears that it is going to run longer than its deadline. The deadline is 10:05 pm this evening and it is now 3:30 pm, but it is not running high priority. It has been running a little over 37 hours and has 0% completed and no time to go. Should it be aborted, or should I let it continue?

I don't have a problem with long running WUs, but if this one is just wasting time, would make sense to stop it and allow other Harmonious Trees WUs to run.

Hello,

Please let it run, in fact we very need long ones to be finished. You can also check out the workunit page to see if your workunit is finished by others or not, to decide whether abort or continue.

If the condition permits, can you please post the ckpt.txt file here?

fwjmath

[DoctorNow]

Damn, I also caught one of this _R_ units. I didn't know before what that was and why it already did run over 60 hours with only 0,002% progress on my quad where I was sure that the limit was 48 hours. I already thought it could be a faulty unit and was nearly to cancel it.
After reading this thread here I hope, it won't run forever...

[Ananas]

... I hope, it won't run forever...

218720 seconds is slightly less than forever - but close