Anybody want to take a guess?

[wolfemancs]

So looking at:
http://www.seventeenorbust.com/secret/
the Dropped-tests queue is down below 300.

I thought it might be fun to try and guess the date when that queue will be exhausted (when the one with the value of n=15999785 gets handed out). It should be right around this same time that the min n for First Pass on the same page jumps almost 900,000 as the values under 16,000,000 are exhausted.

I'm going to take my guess as 09/10/09 (September 10th) at 19:00:00 UTC

CW

PS . . . What tests will the old client start running once the sub 16M tests are gone? Will they just join the new client on the First Pass up around 17M where the new client was going to start being extra useful in it's ability to factor, or will they drop down and work on some of the Second Pass?

[Matt]

September 15th 12:00:00 GMT

[enderak]

I'll go Sept. 13th, 13:00:00 GMT

Personally, I'd like to see them go to work on second pass, but I'm not sure what Louie has planned.

[imwithid]

Out of curiosity, does that mean that all tests under n=16m will have gone through a first-pass (including those old-tests, some of which date back years)?

I'll place my chips on 11-09-09, 08:55.

I've always taken on superstition straight into it's eyes.

Edit: I'd second the second-pass idea for the old SoB client.

[jimmy]

Sept. 11th, 17:21:00 GMT

[jjjjL]

Personally, I'd like to see them go to work on second pass, but I'm not sure what Louie has planned.


I don't know what Louie has planned either. That's a good question. Hmm.... I think if I did nothing it would start handing out the higher tests which we don't want.

So I've set the legacy server to second pass effective now. But I also moved the remaining 420 unassigned tests < 16M into the second pass queue. I tweaked those tests priority so they will be handed out ahead of the actual second pass tests. It's a bit of a hack but I think it accomplishes my goal of segregating the test segments between the old client and new at the 16M dividing line. Well, without me having to sit and watch the server for when that last < 16M test actually gets handed out.

So my solution has actually kind of ruined your game. The correct answer for when the first-pass queue # will jump is now " 4:41 am EST, 29 Aug. 2009 ".


first-pass 48311 16760547 19999951 93 +847855

:)

Cheers,
Louie

[imwithid]

It's as if someone opened up the jar of jelly beans and took out all but one. There goes the lottery.

[shauge]

http://www.seventeenorbust.com/sieve/next.txt does still show 15.9M something.

[enderak]

Looks like that's fixed - it's showing 16762268 now.

[Matt]

I'm a bit confused, was the initial guessing not based on the dropped tests queue? This is currently showing 269 tests for me - is that correct? That works out about a week of crunching left (assuming none more are added). Do we know at what rate tests are dropped and end up in that queue?

[bigmanwhs]

There seems to be something up with the priority of the old client. The number of dropped tests is going up and the Min n is staying at 14616884. I'm not sure what is being given out since I've updated all of mine.

[jjjjL]

There seems to be something up with the priority of the old client. The number of dropped tests is going up and the Min n is staying at 14616884. I'm not sure what is being given out since I've updated all of mine.

I think this is just due to the way I modified how the legacy server to finishing up the last few hundred first-pass tests that when n <16M that I had to be moved into the second-pass queue for logistical reasons.

These special "second-pass" tests are getting handed out and intermediate results are coming in for them so the old server looks like it's doing good work. It's just handing out those specially re-branded first-pass tests ahead of actual second-pass work or dropped tests because the special first-pass tests in the second-pass queue have artificially high priority in the server.

Things will proceed as normal in about 2 weeks when all those special tests get run out of the second pass queue. The legacy pool of clients will then drop back a few million n and start doing real second pass work for awhile and the new GIMPS pool of clients will continue on into the n-higher first-pass tests.


Cheers,
Louie

[wolfemancs]

So we've been crunching second pass again for about a month now, any word back on how our error rates are looking, or any other interesting news from the double check front?