Live info + questions about doublecheck
 

Hello :)

Is it possible to get a "live info" webpage for the doublecheck progress aswell? Something similar to [URL]http://www.psp-project.de/llrnetstats.php[/URL].

How is the doublecheck making progress? Have you seen many instances of doublecheck results not matching original results?

Just curious on the progress being made. :)

Here you are:

[url]http://www.psp-project.de/llrnetstatsdc.php[/url]

Nice, thanks! :)

I would be nice to have a link to it from the main webpage... [URL]http://www.psp-project.de/[/URL]. So I don't have to remember the exact URL. ;)

I'm still curious about what the percentage of mismatches between first run and double check results are.

First thanks for the heads up about the missing link to the double check stats.

About the error rates:
In the region we are double checking at the moment the error rate is surprisingly low. In the region above 3.05M for k=168451 so far we have an error rate of less then 0.5% so far. For the region below that there is no easy answer about the error rate due to the fact that some of the tests expected to be done with llr in reality were done with prp or an older version of prp which create differnt no comparable residues. So when there was an error we only made a third check with llr and not a fourth one with prp. But that would be needed to give a clear answer if it was a wrong residue or only a different application running.

In the lower ranges we also found some error patterens from a user. When this was the case all his results were tested until the error disapeared.
(This happend to five differnt users. I did not contact them as they are not active anymore)

Thank you for the update! :)

So if I understand correctly, there are two equal results produced (out of 3 or 4 tests) for every possible k+n candidate up to where doublecheck is currently?

An error rate of less than 0.5% is quite good I guess. It means the chances of having missed a prime is quite small.

Please inform us on the forum if you feel we should redirect some of our power towards doublecheck. :)

Double checking for all k is at n=1.45M at the moment.

Due to the very low error rate we decided that double checking should concentrate on only one k and try to bring it up to first pass limit.

This has the advantage that we need less resources and at the same time we still see if there is a change in the error rate.

I would like to see three or four more cores run the DC effort to bring it to 5M as fast as possible. I am very interested to see if there is a significant change in the error rate above ~4M.

At the moment most of the work is done by Joe_o and vjs. I have also some manual reservations runing on three cores that have no internet connection.

[quote=ltd;151511]I would like to see three or four more cores run the DC effort to bring it to 5M as fast as possible. I am very interested to see if there is a significant change in the error rate above ~4M.

At the moment most of the work is done by Joe_o and vjs. I have also some manual reservations runing on three cores that have no internet connection.[/quote]

Added 4 cores to DC. :)

Thanks opyrt,

That quad core will help out a bunch, since I lost my quad core Joe and I have been struggling.

J

No probs. We wouldn't want DC to fall too far behind. :)

We did a leap forwards, but it's easy to see from the stats at [URL]http://psp-project.de/llrnetstatsdc.php[/URL] that the [B]n[/B] is getting higher quickly.

Hey opyrt,

ASSUME (probably true but still an assumption) that all k have equal error rates.

and your right but thats also the point of the one k approach.

By doing just one k and bringing it up to the firstpass level (or very close) we get a represenative snap shot of what the actual error rate is for that k.

That k's error rate should represent the error rate of the entire project.

What the approach should show quickly is the following;
- particular users with high error rates.
- if error rate is high in a particular region
- bogus results


Lars will and has been able to identify such high error rates and attribute them to some cause. Although the error rate is very low one can see patterns. By doing these patterns we can be more effective in our doublecheck effort.

hope this helps.