|reply to WireHead |
Re: BOINC I want for christmas
Hmm, then the servers works as they should, the normal progress of wu in seti@home for BOINC is something like this:
1; A wu is sent out to 3 different users.
2; If 1 or more result is reported as download/client-error, or isn't returned before the deadline, an extra copy per error/not returned is given out to yet another user.
3; Then atleast 3 "success"-results is returned, all "success"-results is tried validated.
4; If not atleast 3 of these results is "similar enough", more copies of the wu is distributed.
5; Then atleast 3 results is "similar enough", they passes validation, and one of them is chosen as "canonical result", and all that have passed validation gets the same credit. At the same time the "canonical result" is copied to the seti@home science-database.
Any result that didn't pass this validation is marked invalid, and no credit given.
6; If there's still more copies of a wu currently not returned, and still haven't reached their deadline, BOINC must wait for these results.
Then returned as "success"-result before the deadline, this is only compared to the "canonical result", and credited if similar enough.
7; Then either all results is returned, or all outstanding has passed their deadline, the file_deleter removes all result-files & wu from the BOINC upload/download-directories.
Any after-their-deadline-results that is returned after file_deletion can therefore not be validated, and therefore no credit.
BTW, #6 will only happen in seti@home if a user returned after his deadline, but before the newly distributed copy is returned.
Because of #5 ensuring atleast 3 "similar results", this means also users tested by #6 is indirectly verified against atleast 3 other results. This ensures high probability everyone crunching "correctly" will pass the validation, since being paired off with either 3 cheaters getting the same result or 3 hardware/software-errors managing to get the same wrong result is very unlikely. Therefore point A as mentioned in previous message shouldn't be a problem for normal crunchers.
So for point B & C:
To make sure a wu doesn't have a permanent error so is re-distributed infinitely many times, BOINC is using limits on how many errors is allowed before giving up. There's also limits on #4, and on the total number of results for a wu. Then any of these limits is reached, the wu is terminated and never validated.
Point C is "Then everything is working as it should", and it's unlikely you will successfully crunch a wu that everyone else is having problems with.
Point B is then nothing is working as it should, and another name for this is "the Snap 'snapped' again"...
Well, there's also been some software-bugs crawling around, but AFAIK the server-bugs is fixed and most client-bugs is things like screensaver doesn't work on NT4 so...
BTW, "just guessing there will be a WU, guessing it'll crunch and guessing it'll be returned?" Well, this really isn't any different from any other project like seti@home "classic" have run before so... Especially the months before they got the Cogent-link it was impossible to get wu & return wu. As for the crunching, stumbles over the ocassional "turbo-wu", but AFAIK haven't had any crunching-errors since beta.
"I make so many mistakes. But then just think of all the mistakes I don't make, although I might."