The issue is possibly that a restart writes a few extra 'resume' bits. The validation process looks e.g. at a hash number to ensure all copies, each representing 1/10, make up the total as if a single result was received. If the hash deviates from the majority, that copy is resubmitted until it matches. If too many of the original set deviated, the whole set is resubmitted for 2nd opinion (looking at the or and times of a larger distribution set).

It's not that simple as even clean runs, per the result log, have turned into 'invalid' on a few done on my P4HT. As running in service, never got to see the graphics, so the observation report of the line graph certainly interesting.

Maybe knreed or the other techs could look at the invalids and see if the none critical bits are causing that anomaly?

cheers

Hello bijoalex,
I am sure the techs will do whatever they think is needed. But I have recently heard that the grad students will be working on their doctoral dissertations for this project after we return cycle 27 - the final 2 week period of the year. So I expect that AfricanClimate@Home will go on hiatus for an extended period until the next set of grad students take up the research project and refine the program in light of the initial results. So the time factor is probably not very important.

Lawrence

Seems that something changed in the validation process.
My above mentioned WU ach1_23_48_9-- has got a valid result now. It was on "inconclusive" after the first 10 results came back (+2 for errors) and another set of 10 WUs were sent out.
At the moment we are at 27 (!!!) replications but with astonishing 23 valid results (2 errors, 2 still in progress).

Greetings

Thorsten

The image i have is that if first 10 without raw error are back, a test is applied and e.g. if 8 agree on the control hash, 2 will be send out for recomputation. Eventually if too many do not agree i.e the set is inconclusive, the whole set is resubmitted in one go. Indeed after return of those 10 it might come out that in fact the original majority was valid all along.

Anyway, the admin advised that the next phase will be done differently. The world of bandwidth may look upgraded again by end of this year or 2009 so the whole split might go different, similar to CPDN where e.g. large units are send out and trickles are returned.... nothing has developed how to approach the validation and distribution methodology for the next step.

Not really. Seen this a number of times in past where in the end a majority gets validated. As said the AC@H future is going to see a different way of distribution.