Hello Dirk Gently,
This bug has been fixed by calling TerminateProcess() rather than by calling Exit() in the application. There was a race condition between threads that would sometimes cause problems. BOINC 5.4.9 now will wait until the application is properly terminated before proceeding. As far as I know, there is no reason to select 'Leave in Memory . . .' and plenty of reasons (memory capacity especially) to NOT select it.


The status definitions are here: http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=6105
Since we are using the Quorum of Three method, there must be at least 3 identical results before we define that to be the Valid result. Therefore, if your result is Valid, it matches. If it is Invalid, it does not match. Inconclusive means that we got back at least 3 results, but could not choose a Valid result, so we have sent out more copies of that work unit.


Good thinking. We will want to get a lot of statistics on errors before we broach this possibility. 'I tell you three times' is a very safe rule but might not be statistically justifiable.

Lawrence

On point Dirk Gently's point 2 query and Lawrence reponse "...plenty readon not...", observation is that if left in memory and computer crashes or is booted, the WU very likely gets corrupted and is closed. Zero score. When not left in memory, on restart it picks up from the last datapoint.....you loose several, seen up to 7 minutes of crunch time....anyone should be able to live with that. Anyway on all the validiation argumentation, i've been receiving a few that was of this new 4th copy submission type, with 2 valids on them and the 3rd not having send in anything or an error: If i see these, i freeze the one in progress, but only if that not already had 2 in PV. The 4th copy WU is pulled forward in the queu awaiting its turn, then the 0 or 1once are released again.

I speculate, that the shorter you set the contact time, BUT NOT LESS THAN THE ORIGINAL DEFAULT OF 0,1 days, you will more likely get a 4th unit send, since if deemed reliable, the result would be returned quickest. ONLY, when you plan to go away for a longer time and take your workhorse offline, like i do, do you set the contact parameter to however long the away is going to be. That said, if the 4th copy is send out after 4 days of silence, its pretty pointless to store more than say 5 days worth, since the probability of any older copies already having a quorum of 3 is great, the 4th becoming a academic waste (lest you're just out for the points). I'd say 99,9% of he crunchers are in an area where ISP is up 99,9% of time and the powergrid is reliable by a similar percent. So, knowing a few that like to store a weeks worth, valid reasons excluded, they're doing the cause no favour.....they're just point hungry and like to stand out in their team once a week and force WCG to send out the 4th copy)!

Should this 3 week validity then not be reconsidered? NO, for those that have slow machines or have WU's that do really take long time, it would not be fair.......does the distribution system consider this i.e. even fast machines taking a week and not sending out a 4th/5th/6th copy (I have one that now got 8, including 2 errors).....i suppose one could tweak it out or accept it as some redundancy.

Saturday morning A-life and 1 Doppio Espresso down...enough WCG for the day.

Have a Nice extension of the long weekend