Page 1 of 1

WU Shortages - Prevent cherry picking

Posted: Tue Jun 29, 2010 11:06 pm
by theteofscuba
I'm not happy when I see WU shortages due to people cherry picking the work units that offer the most points or are looking for a specific project.

I noticed that in the bonus point program, you have to ensure that you return at least X% of the work units issued to your account to get bonus points. Could it be possible to use a similar variation of this logic with regards to ALL issued points? -- I'm thinking that if you don't return Y% of issued work units then you should be penalized by either not giving any points or reducing the amount of points earned substantially so that the number of points received will be lower than any work unit that you try to download. I want reduced point earnings penalty to make cherry picking undesirable.

Re: WU Shortages - Prevent cherry picking

Posted: Wed Jun 30, 2010 2:03 am
by ikerekes
Excuse me for being so thick but I don't get it.
At the moment my 3 GPU3 clients are without work for the last 8 hours.
They are all waiting for server 171.67.108.31 which is up and accepting according to the server status page.
I would call the situation typical WU Shortage.

Could you please point me to the Bowl of cherry I could pick from?
I wouldn't even mind if I would be assigned a dreadful 353 pointer even if it producing about 2000 points more than the GPU3 611 pointers. [/end of sarcasm]

Re: WU Shortages - Prevent cherry picking

Posted: Wed Jun 30, 2010 4:08 am
by theteofscuba
PG has said that many times that some WU shortages are caused by people cherry picking. once a WU is abandoned, that WU won't be put back into the queue for a few days at the least or a week at the most, so abandoning many WUs (until you find one you want to run) justs means there are that many fewer WUs available to other people until they expire and are re-released to the queue. couple that with a practice done by many people, that equates to a lot of WUs that could be given to willing and capable donors but can't, because the cherry pickers have abandoned all the WUs that could have gone to a "good" donor.


maybe the better solution could be as simple as allowing users to be honest and simply request to abandon a WU in a way that doesn't harm progress.

Re: WU Shortages - Prevent cherry picking

Posted: Wed Jun 30, 2010 5:49 am
by bruce
theteofscuba wrote:maybe the better solution could be as simple as allowing users to be honest and simply request to abandon a WU in a way that doesn't harm progress.
I've seen some pretty strong opinions that there should be some penalty whether or not the sever gets informed of the abandonment. (Perhaps a lesser penalty if they're honest than if the leave it up to the server to discover it.) How would you feel about that?

Re: WU Shortages - Prevent cherry picking

Posted: Wed Jun 30, 2010 6:21 am
by PantherX
bruce wrote:I've seen some pretty strong opinions that there should be some penalty whether or not the sever gets informed of the abandonment. (Perhaps a lesser penalty if they're honest than if the leave it up to the server to discover it.) How would you feel about that?
Interesting but there are some questions about it:

1) If there was this option to choose the higher PPD WUs by dropping the lower PPD ones, and you were honest and got a lesser penalty, would that be enough to lower your overall PPD or will it be covered-up by the higher PPD WUs your machine(s) is crunching?

2) What if a donor gets assigned a WU without any preference but couldn't make it in time due to DSL problems or hardware issues? Can the Server differentiate between the WU being dumped or a donor ran into some unforeseen problems? If the Server assumes a dumped WU, the donor gets a higher penalty and later when the donor explains their issue, will that penalty be removed?

I think that an easier way around this would be to categorize the Projects in what diseases they are for. Projects that need to be processed quickly can be given a PPD boost while the not-so-urgent ones can have the lower PPD. In this method, the PPD of WUs will be dynamic and would be up to the donor to decide if they would like to contribute to ALS, Huntington's or would like to get high PPD regardless of what Project the WU belongs to.

Re: WU Shortages - Prevent cherry picking

Posted: Wed Jun 30, 2010 6:22 am
by dimilunatic
bruce wrote:
theteofscuba wrote:maybe the better solution could be as simple as allowing users to be honest and simply request to abandon a WU in a way that doesn't harm progress.
I've seen some pretty strong opinions that there should be some penalty whether or not the sever gets informed of the abandonment. (Perhaps a lesser penalty if they're honest than if the leave it up to the server to discover it.) How would you feel about that?
It's not always about abandoning a WU. Sometimes a severe crash like a bluescreen can damage the WU files and the client will go on and delete everything. It has actually happened to me some time last week. How will the server distinguish a donor who just plainly abandoned a WU from a donor who faced a technical issue and had their WU removed?

Re: WU Shortages - Prevent cherry picking

Posted: Wed Jun 30, 2010 6:35 am
by theteofscuba
dimilunatic wrote:
bruce wrote:
theteofscuba wrote:maybe the better solution could be as simple as allowing users to be honest and simply request to abandon a WU in a way that doesn't harm progress.
I've seen some pretty strong opinions that there should be some penalty whether or not the sever gets informed of the abandonment. (Perhaps a lesser penalty if they're honest than if the leave it up to the server to discover it.) How would you feel about that?
It's not always about abandoning a WU. Sometimes a severe crash like a bluescreen can damage the WU files and the client will go on and delete everything. It has actually happened to me some time last week. How will the server distinguish a donor who just plainly abandoned a WU from a donor who faced a technical issue and had their WU removed?
sometimes it isn't the donor's fault, and it should be a rare occurrence. rare enough that failed WUs will not add up to a significant amount to which penalties should apply to. the client could notify the server of the crashed WU in effort to put the WU back into circulation as soon as possible.

Re: WU Shortages - Prevent cherry picking

Posted: Wed Jun 30, 2010 6:50 am
by bruce
WUs which face uncontrollable situations are not considered cherrypicking. They'll be relatively rare, and the server deals with them. It's a whole different issue if somebody intentionally dumps them. Software cannot tell the reasons a WU gets lost, but it can tell when it's a regular occurrence or something that happens rarely. Any policy would have to take both cases into consideration and be fair to everyone. Good luck on that.

A few lost WUs doesn't produce shortages, but when you add more lost WUs which were intentionally deleted, things can get bad rather quickly.

Re: WU Shortages - Prevent cherry picking

Posted: Wed Jun 30, 2010 9:13 pm
by theteofscuba
bruce wrote:
theteofscuba wrote:maybe the better solution could be as simple as allowing users to be honest and simply request to abandon a WU in a way that doesn't harm progress.
I've seen some pretty strong opinions that there should be some penalty whether or not the sever gets informed of the abandonment. (Perhaps a lesser penalty if they're honest than if the leave it up to the server to discover it.) How would you feel about that?
honesty is a good policy. i think that the people specifically looking for points are handled with such penalties. people abandoning for specific "projects" is a bit more difficult to handle. i have doubts that PG would be enthusiastic about allowing folders to specify project preferences when being assigned WUs.

one thing came to mind and that is that the penalties should decay over time to give them a better chance at returning to a normal, healthier amount of points being earned in the future.

edit:
my final concern is malicious users who are requesting WUs with the intention of causing WU shortages. any thoughts on that? Seems like it can be done with some statistical analysis.