Hello!
Some of my results which were normally validated got very low credit values:
260533, 259182, 256869 and some other.
What\'s the reason about it?
/ 
Reply
Quote
http://mindmodeling.org/beta/result.php?resultid=262056
http://mindmodeling.org/beta/result.php?resultid=262836
The validator has not grant that, max errors not reached.
/ Reply
Quote
There is a cpu time problem on linux machines, so 0sec=0 credits (min quorum 2 is always 0 too)
The second issue is that a WU is sending out up to 3 times to the same host!
/ Reply
Quote
I have a WU with 0 credits but it hasn\'t errored out, hmmm...
See http://mindmodeling.org/beta/result.php?resultid=257580
____________
Greetings from Cori
/ Reply
Quote
It looks like credit is based on the first valid result returned. If the first returned is linux, it gets credited with .001, and then all subsequent valid results also get credited with .001, regardless of claim. Until this is fixed, the linux (and Mac) apps should be removed/pulled.
1) Obviously the Linux problem needs to be fixed, so that they are awarded credit. Also so that they do not drag down the other wingmen.
2) I think the credit award method should change to something like the average of the first two, since a quorum of 2 are needed.
3) If a quorum of 2 are needed, an initial replication of 5 is *way* too many. IMO, an initial replication of 2 is what it should be set to. Bump it up to 3 if the project needs validated results quickly.
____________
Dublin, CA
Team SETI.USA
/ Reply
Quote
Jack.HarrisVolunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 24 Apr 07
Posts: 499
Credit: 626,319
RAC: 33
It looks like credit is based on the first valid result returned. If the first returned is linux, it gets credited with .001, and then all subsequent valid results also get credited with .001, regardless of claim. Until this is fixed, the linux (and Mac) apps should be removed/pulled.
1) Obviously the Linux problem needs to be fixed, so that they are awarded credit. Also so that they do not drag down the other wingmen.
2) I think the credit award method should change to something like the average of the first two, since a quorum of 2 are needed.
3) If a quorum of 2 are needed, an initial replication of 5 is *way* too many. IMO, an initial replication of 2 is what it should be set to. Bump it up to 3 if the project needs validated results quickly.
Good Observations.
First in wins
-- This is the default/sample that comes with BOINC
-- I will try to fix this soon to something similar to what you suggest
I will look into the error you describe in #1
--Jack
/ Reply
Quote
Thanks for looking into it. ;)
Btw: I\'ve noticed that with WUs where I was granted 0.001 credits the WU was sent not only to a low-claiming Linux host, too - that Linux host got the same WU four (!) times! 
And similar happened with me and a Dual Core as wingman. The Dual Core crunched the WU three times. And once he got such low credits (0.91) that it ruined my claimed credits... ;-)
See: http://mindmodeling.org/beta/workunit.php?wuid=48537
and: http://mindmodeling.org/beta/workunit.php?wuid=48554
____________
Greetings from Cori
/ Reply
Quote
It looks like credit is based on the first valid result returned. If the first returned is linux, it gets credited with .001, and then all subsequent valid results also get credited with .001, regardless of claim. Until this is fixed, the linux (and Mac) apps should be removed/pulled.
1) Obviously the Linux problem needs to be fixed, so that they are awarded credit. Also so that they do not drag down the other wingmen.
2) I think the credit award method should change to something like the average of the first two, since a quorum of 2 are needed.
3) If a quorum of 2 are needed, an initial replication of 5 is *way* too many. IMO, an initial replication of 2 is what it should be set to. Bump it up to 3 if the project needs validated results quickly.
Good Observations.
First in wins
-- This is the default/sample that comes with BOINC
-- I will try to fix this soon to somethings similar to what you suggest
I will look into the error you describe in #1
--Jack
Thanks. I just re-read my post, and it comes off a bit harsh. Truly, I was not trying to demand anything, but rather just putting my opinions out there. Sorry for the less-than-pleasant tone.
____________
Dublin, CA
Team SETI.USA
/ Reply
Quote
Btw: I\'ve noticed that with WUs where I was granted 0.001 credits the WU was sent not only to a low-claiming Linux host, too - that Linux host got the same WU four (!) times!
I believe there are three levels within the BOINC server, regarding results for the same WU:
1) multiple results from the same WU can be issued to the same machine
2) multiple results can be issued to the same user, but only one per each of his machines.
3) only one result per user, regardless of the number of machines.
It looks like we may be at #1 now. Usually, projects go with #2 at a minimum. This prevents a faulty box being able to validate a faulty result against itself. #3 is unnecessarily restrictive, IMO.
____________
Dublin, CA
Team SETI.USA
/ Reply
Quote
Btw: I\'ve noticed that with WUs where I was granted 0.001 credits the WU was sent not only to a low-claiming Linux host, too - that Linux host got the same WU four (!) times!
I believe there are three levels within the BOINC server, regarding results for the same WU:
1) multiple results from the same WU can be issued to the same machine
2) multiple results can be issued to the same user, but only one per each of his machines.
3) only one result per user, regardless of the number of machines.
It looks like we may be at #1 now. Usually, projects go with #2 at a minimum. This prevents a faulty box being able to validate a faulty result against itself. #3 is unnecessarily restrictive, IMO.
Well, but in these cases I mentioned it was not so good because a) every Linux WU failed or b) the outcome was different although the same machine crunched the WU three times.
And I got always very low credits by that, too. ;-(
____________
Greetings from Cori
/ Reply
Quote
I get too much WUs with too short deadline, so every other project is stopped and the MindModeling WUs are all running with high priority. So I only can stop downloading WUs fom MindModeling.
Further the granted credits are too low, under 10 per hour on a Pentium D 3.00 GHz.
/ Reply
Quote
3) If a quorum of 2 are needed, an initial replication of 5 is *way* too many. IMO, an initial replication of 2 is what it should be set to. Bump it up to 3 if the project needs validated results quickly.
Good Observations.
First in wins
-- This is the default/sample that comes with BOINC
-- I will try to fix this soon to something similar to what you suggest
--Jack
Other projects have often started with a high initial replication until they gained confidence that an initial replication of two works well, which utilizes a donor\'s computer resources efficiently and efficiently moves the project forward.
The heuristic projects like Rosetta and QMC only require one result because of the probabilistic nature of their calculations, while projects that use deterministic calculations such as Einstein@home are satisfied with an initial replication of two.
From my standpoint, I believe in the most efficient use of computer resources (including time and the cost of energy and reducing the carbon footprint), and so I heartedly support an initial replication of two! :)
____________
Regards,
Bob P.
/ Reply
Quote
Double post, sorry, and I wish BOINC allowed a way to delete these!!
____________
Regards,
Bob P.
/ Reply
Quote
Btw: I\'ve noticed that with WUs where I was granted 0.001 credits the WU was sent not only to a low-claiming Linux host, too - that Linux host got the same WU four (!) times!
I believe there are three levels within the BOINC server, regarding results for the same WU:
1) multiple results from the same WU can be issued to the same machine
2) multiple results can be issued to the same user, but only one per each of his machines.
3) only one result per user, regardless of the number of machines.
It looks like we may be at #1 now. Usually, projects go with #2 at a minimum. This prevents a faulty box being able to validate a faulty result against itself. #3 is unnecessarily restrictive, IMO.
Well, but in these cases I mentioned it was not so good because a) every Linux WU failed or b) the outcome was different although the same machine crunched the WU three times.(
To make the results ensured the system of quorum is used, I think.
suguruhirahara
____________
/ Reply
Quote
Jack.HarrisVolunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 24 Apr 07
Posts: 499
Credit: 626,319
RAC: 33
A new results validator has been added to MM@Home
Credit is now given in the following way.
After 2 valid results are returned for a workunit, the average of the 2 credits requests are given
Valid = proper return data structure etc
Now that validity of results are checked the averaging of good work with broken results(.001 run-times) no longer occurs. This should help strange granting values from occurring.
--Jack
/ Reply
Quote
Credit is now given in the following way.
After 2 valid results are returned for a workunit, the average of the 2 credits requests are given
The average is not being given, it\'s still granting the lowest request. Here are some examples:
http://mindmodeling.org/beta/workunit.php?wuid=263736
http://mindmodeling.org/beta/workunit.php?wuid=264165
http://mindmodeling.org/beta/workunit.php?wuid=266407
http://mindmodeling.org/beta/workunit.php?wuid=264305
http://mindmodeling.org/beta/workunit.php?wuid=264214
http://mindmodeling.org/beta/workunit.php?wuid=275172
/ Reply
Quote
All mine are still giving the first one in, which I am not that one, and not the average of the two. Its a struggle when a WU takes over an hour to complete yet I only get a few points for it.
Ni
/ Reply
Quote
Good point. it looks like the credits are still be granted based on first in, rather than the average of the quorum. Can this please be changed?
____________
Dublin, CA
Team SETI.USA
/ Reply
Quote
It\'s not granting first in, it\'s granting the lowest. On this wu I was the first to report but I was granted the lower amount.
/ Reply
Quote
It\'s not granting first in, it\'s granting the lowest. On this wu I was the first to report but I was granted the lower amount.
Ouch! That\'s even worse!
____________
Dublin, CA
Team SETI.USA
/ Reply
Quote