question about secondpass, error rates and the graph...

[vjs]

There has been some discussion on our team forums regarding the number of secondpass tests error rates and "The graph" posted a couple weeks ago and some additional comments. I thought I'd go to the source and ask for comments and clarification...

We were just wondering if there were going to be any updates or if additional information has been gathered.

First several people have noticed that the number of secondpass tests now exceeds the number of firstpss tests completed per day or they are now roughly the same. We have been putting our slow machines onto secondpass, while encouraging those with faster machines to continue as they were.

1.

Some questions remain, since the number of tests per 1M n are about the same for 1M and 8M

1M<n<2M has 26903 tests
and
8M<n<9M has 26557 tests

yet looking at the trends over the past 10 days...

1st pass queue Min-n has increased 11,323.4 a day.
While the 2nd pass queue Min-n has increased 7,877.6 a day

We are curious about this... have factors not been applied, are we testing all n's less than the k/n prime (we havn't noticed prime k's being handed out), is the residue match showing alot of mismatches, are doublechecks handed out for a third time due to slow clients the 24 hour? Should this be increased?

2.

Were comments going to be made regarding the large graph, answers posted?

3.

Do we have a decent analysis of the error rate? Or are the different clients producing different residues?

4.

Has the optimal n-level for secondpass been determined yet or are we currently sticking with the secondpass n should be half that of first pass n.

5.

Is there a progress report for V3, and have all the server issues been ironed out?


Thank-you very much please answer what you can, we understand that all of the answers are probably not available.

Thank-you for your time.

[ShoeLace]

the answer to (1) is PROBABLY that 1M-2M test take considerably less time..

such that even with less machines 'assigned' the actuall test thruput is greater

eg.. a 1.5 M test takes 4 hours on my machine and a 8M test takes nearly 2 days

so you only need say 1/10th as many machines to get teh same low n thruput.

I dont know the answer to your other questions.

[vjs]

Shoelace,

Actually the point of the question is that with about the same number of tests completed, first pass has managed to advance 30% further than secondpass. This is not in the dailys total (dropped tests etc) but in the assignment level over a 10 day period.

i.e. from 8,000,000 to 8,011,325 compared to 1,300,000 to 1,307,878

[MikeH]

I suspect that part of the aparent differences are due to the drop-tests queue. The number of main PRP tests being completed is a sum of first-pass and droped-tests, the DC is only the second pass queue.

If you take a look at the sieve project stats (http://www.aooq73.dsl.pipex.com/2005/scores_p.htm), the 1 and 14 day averages of the two 'issued' columns probably give a better (or at least different) indication of what you're looking for.

We can also see from the DC tests to be issued 1 and 14 day averages that the DC is (just) out running the main effort (number -ve, currently -3), but over the last 14 days, main PRP has been running well ahead of DC (+ve, currently +70).

So it looks to me that there's nothing suspect going on here, no candidates of prime finds being tested etc., just different ways of displaying information.

I'd be very interested to see updated error rates.:cheers:

[Nuri]

Originally posted by MikeH
I'd be very interested to see updated error rates.:cheers:

:bouncy: