Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Project Staging Area :
New PRPNet Policy on Bad Results
| Author |
Message |
|---|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
                               
|
|
In the interest of fairness, we are implementing a new policy for PRPNet projects. If PrimeGrid determines that a participant's computers are returning an excessive numbers of erroneous results, we may, at our discretion, suspend the awarding of PSA BOINC credit for that participant. Since PRPNet work isn't normally double checked, erroneous results currently get full credit. It isn't fair to participants who are returning good results.
It is the responsibility of all participants to insure that their computers are working properly. On BOINC tasks, the results are double checked, and bad results won't get credit. But on PRPNet the results normally are not double checked, and it's very important that everyone makes sure their computers are working properly.
EDIT (6/8/2016): In order to protect the integrity of the work being done, we also reserve the right to take more serious actions, at our discretion, including, but not limited to, prohibiting users from using PRPNet.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
I like the new policy.
I think the policy will have an effect in practice only in some PRPNet projects ("ports"). For projects where random calculation errors (due to hardware problems) typically yield a recognizable result, like 0000000000000000, it will be possible for administrators to identify the problem (and suspend awarding of credit).
However, for some projects, the "result" will typically appear "random", like 6D97B073E098E888, both if calculations took place correctly, and if random hardware errors were present. So in the absence of double-checking, bad computers operating (only) in these projects, can be impossible to identify(?).
/JeppeSN | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
|
However, for some projects, the "result" will typically appear "random", like 6D97B073E098E888, both if calculations took place correctly, and if random hardware errors were present. So in the absence of double-checking, bad computers operating (only) in these projects, can be impossible to identify(?).
/JeppeSN
Indeed. I am not at all happy with this scenario. When there's a problem (notice I said "when", not "if"), it's often undetectable. That makes the quality of the results exceptionally suspect. Not only is there some percentage of results that are wrong; you have no idea how large or small the problem is, or even that there is a problem.
It's a big reason why many PRPNet ports have been moved over to BOINC in the last few years. That is a trend that will likely continue.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
Why not just implement double checking on PRPnet?
____________
1437 · 2^495800 + 1 | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
|
Why not just implement double checking on PRPnet?
In theory it's as easy as changing a single line in the configuration file, but in practice it would create a whole new set of problems. One of the larger problems would be that it would start double checking at the beginning rather than just double checking new work. We would need to wipe the database before turning it on. That's something we'd rather not do.
____________
My lucky number is 75898524288+1 | |
|
|
|
start double checking at the beginning rather than just double checking new work.
That might not be a bad idea. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
|
start double checking at the beginning rather than just double checking new work.
That might not be a bad idea.
It wouldn't be a bad idea at all... on BOINC, where there's a lot more computing power available. In fact, when we move a project from PRPNet to BOINC we tend to do exactly that. On PRPNet, it wouldn't work nearly as well since it would take a few years to catch up. People would lose interest pretty quickly.
We also don't have much -- actually any -- experience with PRPNet double checking. There's a lot of nuances involved in comparing two tasks and it would require a bit of research on our part, and likely numerous changes to PRPNet to get it working exactly the way we want.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
We could port everything to BOINC and just use PRPNet for testing and beta work.
Just a thought.
____________
| |
|
Post to thread
Message boards :
Project Staging Area :
New PRPNet Policy on Bad Results |
