1 Week Dealine Long Enough?
9 Jan 2005 15:10:04 UTC
Topic 187220
(moderation:
Is the 1 week deadline sufficient? Given that the new work units take at least 10 times longer to complete the deadline should have been extended by the same amount. Many users participate in other projects as well and can take a while to complete a unit. I have a couple of the new work units that have two users go past the dealine and not return results.
Any comments from the project team?
1 Week Dealine Long Enough?
)
I'm not a member of the project-team, but I think, it should not.
The max queue-deep for the client is 10 days, so, if the client does, what it should, the deadline doesn't need to be longer than 10 days.
Supporting BOINC, a great concept !
> I'm not a member of the
)
> I'm not a member of the project-team, but I think, it should not.
>
> The max queue-deep for the client is 10 days, so, if the client does, what it
> should, the deadline doesn't need to be longer than 10 days.
>
No. That would mean that CPDN would only need a 10 day deadline. The WUs there take one to six months to complete.
The WU deadline needs to be extended to match the computation length, not the queue size.
BOINC WIKI
But then the question is,
)
But then the question is, what is the computation length? It varies by computer. Deadlines must be set based upon "Scientific need". The participants must take deadlines into consideration, when deciding what their computers can crunch. If you have a computer that cannot meet the minumum criteria of a project, then it would be better to select another project.
At some point, participants must make "common sense" decisions about their equipment, and their settings. If you have a slow computer, crunching several projects does not make sense. Settle on 1 or 2, and set a queue that does not download excessive amounts of work.
Ultimately, I suspect that all BOINC projects may need to set "minimum recommended" computer specificiations to meet that projects need.
I know that 6 out my 7 computers could crunch 3 or more projects 24/7 without fear of missing deadlines. (Predictor may be an exception because of very short deadlines). The 7th computer, my iBook 600 is best limited to no more then 2 projects with 7 days or longer deadlines, small queues and probably is best off left crunching only one project (which is exactly what I do).
It would be nice if BOINC could set queues on a project specific basis, whereby downloads could be restricted to no more then (for example) 50% of the shortest required deadline. Maybe someday that capability will be there. For now it isn't, so we must live with the constraints we have, until improvements can be made.
If the it does not harm the science to extend the work unit deadlines for Einstein@Home then I am all for it. However, if the scientific need is for shorter deadlines, than that is what we (the participants) must adjust to. Only the project team can make that decision. I'll live with their decision.
Science first. Credits and competion second. I for one, would rather focus my resources on 1 or 2 projects that I can adequately support, then try to do 4 or 5 projects, and harm the science for some or all of them. If Science does not benefit first, then there is no sense in the project even existing.
Team MacNN - The best Macintosh team ever.
Your point about science
)
Your point about science coming first is right on the money and the science should be on top. The current dead-line is 7 days. For one of my units the, http://einsteinathome.org/workunit/295544, I and another user completed the work in approximately 2 1/2 days each. Well within the limts of the dead-line. The system then waits for the other users to finish. Now, these two users have now gone past the dead-line and the system issues the unit to a new user. We will now wait for this user to finish or not finish the unit. So, from a "science" perspective the work is not complete as the results still cannot be verified based on the current rules. So, we keep on waiting. What if, those two users who did go past the dead-line just needed another day or two? We would have had the result and been ahead of the game.
The real question here is, "How was the dead-line determined?" Did someone just pull a number out of thin air, "...hey, one week sounds good"? Or, was there more thought put into this? Did they look at the distribution of time spent on work units by users and gone for a 95% or greater confidence interval? I don't know. But, if you create work units that take X times longer to complete, then the dead-line should be increased proportionately (BTW, I do not recall what was the dead-line with the previous work units).
>Now, these two users have
)
>Now, these two users have now gone past the dead-line and
So *that's* why my pending credit never gets below about 40WUs/800 rocks...
Cheers,
PeterV.
...We will now wait for this
)
...We will now wait for this user ...
.. the reditribution of this work was downloaded by one of my hosts.
got the dead line for this specific work for 15.01.2005
The crunching host will be an AMD Athlon 2100, needes about 6:30 for this work.
The WU sittes in a queue, so I guess, "your" WU will be done
in about 2-3 from now. Might earlier.
The 2 now "valid" results have generated 82.76 and 102.77.
*my* host will generate about 62.85 (taken the value from other results)
Marcin, watch again in about 2-3 Days;-)
The thing i find odd is that
)
The thing i find odd is that the new longer WU’s have a deadline 3 days shorter then the old short ones (10-7=3). I agree with you that the science comes first, but on the other hand, less computers crunching for your project = less science done. My pIII 450 crunches one of the new longer WU’s in about 40H so 7 days works but a 10 day deadline would give me some margin.
Then you're really interested in a subject, there is no way to avoid it. You have to read the Manual.
In the thread below, Bruce
)
In the thread below, Bruce Allen, comments about the shorter deadline being just for testing purposes. They will go back to the longer deadline when the project goes live.
Thread 191
[img]http://www.boincstats.com/stats/banner.php? cpid=19ffeb205afac117477975798c4ada3f[/img]
> The thing i find odd is
)
> The thing i find odd is that the new longer WU’s have a deadline 3 days
> shorter then the old short ones (10-7=3). I agree with you that the science
> comes first, but on the other hand, less computers crunching for your project
> = less science done. My pIII 450 crunches one of the new longer WU’s in
> about 40H so 7 days works but a 10 day deadline would give me some margin.
>
Hi Ziran,
remember we're still in test (and yes, Guido, we are, so please try and keep as
much patience as you can, pretty please).
Workunit deadlines have to be carefully adjusted, as have to be maximum file
sizes (both for download and upload), and we're also still working on server
code a lot (that's why you may get fewer responses these days).
S
> The thing i find odd is
)
> The thing i find odd is that the new longer WU’s have a deadline 3 days
> shorter then the old short ones (10-7=3). I agree with you that the science
> comes first, but on the other hand, less computers crunching for your project
> = less science done. My pIII 450 crunches one of the new longer WU’s in
> about 40H so 7 days works but a 10 day deadline would give me some margin.
>
Hi Ziran,
remember we're still in test (and yes, Guido, we are, so please try and keep as
much patience as you can, pretty please).
Workunit deadlines have to be carefully adjusted, as have to be maximum file
sizes (both for download and upload), and we're also still working on server
code a lot (that's why you may get fewer responses these days).
S