I have made some observations regarding the validation and score of workunits.
1) I have a couple of validation errors with my Linux WUs. Almost all of them (although not entirely) were against two Windows machines. So there might be a cross-validation problem as other projects have experienced as well. On the other hand it might simply be a coincidence because there are more Windows- than Linux machines working on the project.
2) I have the feeling that whenever two Linux machines agree upon a result and validate, the score will be 23.15. If a windows machine is part of the quorum it will be around 1 or less.
Regards,
mickydl*
/ 
I have some marked Invalid against other Windows Box's, in fact some are against my own Box's. One Box with give an Invalid then another Box of mine will pickup the same Wu & it will be marked as valid ... Weird
/ 
The moral of the story is that everyone should run Gentoo Linux. I was wondering about the invalids. All hosts I'm using are getting them, also. I was hoping it wasn't because of Linux.
/ I got some comments about this yesterday. At the beginning looked like it happened when the two machines where virtualised, but now I realise that the credits are unfairly high just when two Linux machines match, up to 20 times more per a single WU than the Windows host.
I'm stopping the project a few hours until I can fix it. Otherwise credit scores will have no sense.
Sorry for this.
/ I have checked the data base. There are circa 78200 WU processed, and 6% of them have an average canonical credit of 22.4, while the 94% have an average canonical credit of 0.72. This produces a big unbalance in the results, because this 6% produces more than twice credits than the other 94%.
20 credits is more or less the credit given for a WU able to last its simulation to the end (a couple of hours), but not for the 2 minutes they are taking on average. Don't know why Linux machines are claiming that, something wrong in my config files I guess.
This and the validation issue has to be investigated but will take long. In the meanwhile I can set a work out for this. The credit value will be filtered, so any value higher than, say 2 credits, will remain as 2.
Also, now I'm not sure what could I do to fix this in the user's credit and results tables. Users with only Linux machines have chances of having more credits that they should have compared with users only with windows machines. A little mess. Perhaps the best solution is just let the project run with this fix and let the credits adjust as more WU are calculated.
/ Also, now I'm not sure what could I do to fix this in the user's credit and results tables. Users with only Linux machines have chances of having more credits that they should have compared with users only with windows machines. A little mess. Perhaps the best solution is just let the project run with this fix and let the credits adjust as more WU are calculated.
As long as the problem is fixed I wouldn't do anything about the excess credit some/all of the Linux Box's were getting. Nobody's gained that much from it & unless you could run a script to correct the Credits it would be to much work on your part to manually go thru all the Linux Box's to correct it so I'd just let it stand ...
/ Yeah. I also vote that it'd be too much work for you to safely clear up those mis-credited WU's. 8-)
Seriously though, if you can find a quick and clean way to do it right, I sure won't scream "foul" over it. That's what a Test phase for a project is all about. I had noticed the credit disparity, but hadn't realized it was just Linux-Linux validations until my team-mate pointed it out to me last night. I thought there was some specific reason those WU's were more valuable. Oh well. Time for my systems to earn their way. :-)
/ Yeah, I figure my opinion would be obviously biased, so I didn't bother to offer one. I'm just hoping to not drop down below 25k. If I don't lose any credit, I'll continue crawling my way to 50k. I'm just kind of waiting to see what happens before I continue.
/ Here's a curious wu.
http://falua.cesfelipesegundo.com/Neurona/workunit.php?wuid=78101
It finished quickly, validated with a Windows computer and still got 3 credits. I'm wondering what's special about this one that it finished in half the time and got double the credit in spite of a Windows wingman.
/ Have to check the result returned for this WU. For some reason the Windows host requested 2.99 credits. The credit system is pretty weird.
/ 
are the wu's gonna be longer ? langer=more points? :)
____________
Join us at www.boincatpoland.org
/ are the wu's gonna be longer ? langer=more points? :)
If a WU is a match with the searched solution, it should require about 100 minutes of calculations, but so far all WU have been found "bad", and are calculated in a few minutes.
/ 
If a WU is a match with the searched solution, it should require about 100 minutes of calculations, but so far all WU have been found "bad", and are calculated in a few minutes.
Do you know if such a solution can be found by the current application or do you use this run to find it out?
/ If a WU is a match with the searched solution, it should require about 100 minutes of calculations, but so far all WU have been found "bad", and are calculated in a few minutes.
Do you know if such a solution can be found by the current application or do you use this run to find it out?
The current application is able to recognise "good" solutions, but I don't know where is it placed in the solution space. For this reason we are sweeping different combinations.
/ Well, I think the problems are almost solved, but not entirely.
This WU looks suspicious :) Interestingly it's a quorum of three and it gets exactly three times the score than the previous quorum two WUs. All three hosts are Linux. So far I have found only one such WU.
Regards,
mickydl*
/ I thought there was going to be a Limit of 2 Credits ?
/ I thought there was going to be a Limit of 2 Credits ?
It is, I have set up this in the validator, but as in many other BOINC aspects, seems that it is not working <sigh> as it should.
/ No, false alarm, its OK. The value in the column is the claimed credit, but the filter I set yesterday (finally at 3 credits) works:
http://falua.cesfelipesegundo.com/Neurona/workunit.php?wuid=94436
Nombre Prueba081503.pro_1308943625
Programas Neurona@home
Crear 24 Jun 2011 19:27:05 UTC
canonical result 217363
crédito concedido 3.00 -> Credit granted
minimum quorum 3
initial replication 3
Now I think the linux clients are claiming the rsc_fpops_bound I have set. Well, at least is a clue.
/