A year ago, Mike H said December 2004, and I said January 2005.
I'm changing my guess to December 22, 2004.
What is your guess?
Printable View
A year ago, Mike H said December 2004, and I said January 2005.
I'm changing my guess to December 22, 2004.
What is your guess?
I think it will be January 28th, 2005 - unless Louie et al. forget to put new tests in the "first test" queue, that is. :D
1 to 2 000's n tested per million k
369 tests per day
==> 10 days
Only 284 tests have been done for first time tests.
I would stick to the n Increase row - although that deviates due to the dropped tests queue...
So you are telling me that I should have stuck to my mid January guess?
It looks like January 10th, 2005 right now. But we haven't seen the full results of the new client yet. That should kick in soon. Soon enough to bring the date into 2004? Who knows. Keep your guesses coming!
Another parameter to take into account is, if we'll find a prime in between, and if so when (whicn n level) and for which k?
It might bring in an additional discrepancy of up to a month (or even more, in case it attracts some new users)...
Or maybe I trick you to believe this and change back your guess, just to take December 22nd and win the jackpot! :DQuote:
Originally posted by Joe O
So you are telling me that I should have stuck to my mid January guess?
Just kidding - I really think it will take between 1 and 1.5 months. Unless there are run-away clients, of course.
Let's just hope its for a high prime weight k
Example of how many test will need to be performed for each k between 8m and 9m
Sorted by increasing # TestsCode:Tests k
3184 4847
3774 10223
1401 19249
3286 21181
1298 22699
3297 24737
1878 27653
1773 28433
3145 33661
4367 55459
1080 67607
formatting /ceselbCode:Tests k
1080 67607
1298 22699
1401 19249
1773 28433
1878 27653
3145 33661
3184 4847
3286 21181
3297 24737
3774 10223
4367 55459
Thanks ceselb,
how does one format tables etc on this board. is it html ???, etc
Also has anyone noticed this from http://www.seventeenorbust.com/secret:
first-pass 9100 7676950 7999964 229 +11439
Did we just take a huge drop in the number of tests to reach 8m??? I thought it was over 10K yesterday... Could be wrong.
It's the [ code ] tag - I don't remember how to make a tab stop, though.Quote:
Originally posted by vjs
how does one format tables etc on this board. is it html ???, etc
AFAIK, it was already at ~ 9.5k yesterday. 3 days ago, it was > 10K, of course.Quote:
Did we just take a huge drop in the number of tests to reach 8m??? I thought it was over 10K yesterday...
Thanks Mystwalker,
I guess I didn't check that page for a while and going from 5 -> 4 digits is sort of a mental hey look at me I'm decreasing.
I was just hoping that for some UNKNOWN reason we wouldn't have to test approx 1/11th of the pairs :) to get to 8m.
format things with notepad (or any block font tool), then stick it in between [ code] [ /code] tage (without the extra space).Quote:
Originally posted by vjs
how does one format tables etc on this board. is it html ???, etc
code keeps the tabs and spaces intact.
It's really looking like I should have stayed with my January 14th, 2005 guess.
It still looks like January 10th, 2005. After the immediate increase with the announcement of the new client for non SSE2 machines, the rate has held pretty steady for four days now. How long is the "Average" PRP test taking to complete now, as opposed to before? Could it be we haven't seen the full effect yet? Will it kick in just in time for Mystwalker's guess to be correct? Or really get into high gear so that tqft's guess will win? Stay tuned!
Keep your guesses coming!
first-pass 8224 7709804 7999964 264 +12977
(7 999 964 - 7 709 804) / 12 977 = 22.3595592 from calc.google.com
6.01.2005
Well, how about an e-mail announcement for all users? There won't be too hard to send 10k email's but this can gain some users attention. i said about users that set up client long ago and then paid no attention to site of forum.Quote:
Originally posted by Joe O
After the immediate increase with the announcement of the new client for non SSE2 machines, the rate has held pretty steady for four days now. How long is the "Average" PRP test taking to complete now, as opposed to before? Could it be we haven't seen the full effect yet?
and does client report back it's version?
can we see some stats?
Total number of users 8281 users
Users who have ever submitted work 6617 users (79%)
Users active in the last 2 weeks 989 users (11%)
such a pity that only 989 / 82.81 = 11.9430021% of users are active :spank:
maybe we must :Pokes:
and what's with rounding on a stats? 11.94% sounds more like 12% ;)
this is a perl or php specific?
oh well, really i don't care so much, so do what you want :crazy:
Reminder, this is in DD.MM.YYYY format, so for us date challenged, January 6, 2005.Quote:
Originally posted by Death
6.01.2005
So, to recap:
December 25th, 2004 tqft
January 6, 2005 Death
January 10, 2005 Joe_O
January 28th, 2005 Mystwalker
email might work - except a lot of bounces and spam treatment
Post notice to ArsTecnica and Slashdot?
It's looking like January 12, 2005 right now.
December 25th, 2004 tqft
January 6, 2005 Death
January 10, 2005 Joe_O
January 11, 2005 Royanee
January 13, 2005 Keroberts
January 16, 2005 Frodo
January 21, 2005 VJS
January 28, 2005 Mystwalker
Doesn't anyone else want to hazard a guess?
MikeH?
Nuri?
????
So as of December 21, 2004 830PM it now looks like January 6th, 2005.
That's quite a swing in one day!
Probably because Royanee picked the 11th. <G>
O.K. I'll bite
Jan 21 st
I'll also go really out on a limb and say we will find a prime shortly there after by the end of Feb.
A wild guess from here
January 16, 2005 Frodo42
gimme jan 13
hehe, I'll take 1/11 to be mean to you Joe. ;)
Arrrgh,
I forgot to take into account all of those computers everyone is going to borg at grandmas house over christmas, right guys, quick service install, etc....
Oh and hey is this like "the price is right" where you bid one day more???
Then I'll take the 28th Bob, I mean Joe, oh oh oh, 26, no 27...
sorry that's taken,
fine Dec 22 so I don't go over :rotfl:
Guess I'll stick with my earlier, come on it was funny. :jester:
first-pass 6060 7787071 7999964 241 +9653
6 060 / 241 = 25.1452282 days from now - 18.01.2005
another method
(7 999 964 - 7 787 071) / 9 653 = 22.0545944 days from now - 15.01.2005
that was slightly different....Quote:
(7 999 964 - 7 709 804) / 12 977 = 22.3595592
It's looking like January 13, 2005 right now.
December 25th, 2004 tqft
January 6, 2005 Death
January 10, 2005 Joe_O
January 11, 2005 Royanee
January 13, 2005 Keroberts
January 16, 2005 Frodo
January 21, 2005 VJS
January 28, 2005 Mystwalker
Doesn't anyone else want to hazard a guess?
MikeH?
Nuri?
????
I knew I should have stuck to my guns with my original guess from last year!
Keroberts has good shot, but Royanee just might win after all. But Frodo and VJS are still in the running, as is Mystwalker.
Are there any more guesses?
To be honest, it would not be fair to jump in to the competition at this moment, when everything is much more clear when compared to the time most of the guesses were made...
BTW,cool!! :cool:Quote:
Another parameter to take into account is, if we'll find a prime in between, and if so when (whicn n level) and for which k?
And a second BTW, does anybody know/remember what would happen in the case where all first-pass tests are distributed, and for some reason whatsoever, no new tests are dumped to the database? I mean, what if 7999964 first-pass assignment limit is not removed timely? Will the server start distributing second-pass tests to the individual clients? Even if we let that happen just for a handful of days, double check limit will reach somewhere close to 3 million level at no time...
Pros:
- If there is a missing prime at that easy to grab region, we'll be left with average 10% less PRP work going forward.
- It's a good time to squeeze such work, just after a prime is found. Existing users might like the idea of doing something different for a short time?! Also, in case there will be a rush of newcomers, they will start with relatively shorter tests and see results, which might increase new user retention ratio.
- There will be slight (1-2%?) increase in sieve speed due to rapid decrease in range size.
- We will have more concrete data on error rates. Tests for n<1m takes relatively no time, thus very low error rates so far. But, what about some larger n? Do we have sufficient data on that?
- I will not wonder anymore if there was a missing prime in the 2m-3m range.. :D
Cons:
- Ordinary users will wonder what's going on, and some might get angry because of not being told their machines are working on so called useless second-pass tests. More important than that, a lot of them might really get mad because of the cEMs/sec speed decrease they observe on their clients and stats. Still, this will be a one off speed decrease (also for only a limited time), and will apply to all users. Thus, hopefully, relative positions (rankings) will not be affected as much as one might be scared of. Of course, to reduce the extent of potential "riots" one should also announce (at the news page of the site) that second-pass tests will be distributed for, say one week, and explain reasons for that.
- Which brings the issue of, is it worth it? I mean, is second-pass ahead of it's schedule or lagging behind, when compared to the optimum n level it should be when compared to first-pass level? I remember a ratio of 4x-5x that was discussed (was it?), but how accurate is that ratio? In the end, it's a parameter of the error rate.
This might sound like I'm suggesting we do it, but I'm not. Consider this as an idea that came to the mind during brainstorming session... ;)
And, BTW, if we do it, VJS or Mystwalker will be the winner at >8m competition... :D
Nuri,
I always liked the idea of handing out a few double check tests per day into the main que. Something like 100... but I totally agree switching everything to doublechecks would be interesting and I think most users are for it (for a short time)but probably a bad idea.
I would also like the idea of doing it now b/c of the prime everyone is interested and it wouldn't catch people as off guard.
I made a more detailed post in the 2.2 vs 2.3 critical update tread regarding exactly what your talking about.
If it were done this way most people probably wouldn't notice, besides 1.3m tests are being handed out in the main que from time to time.
You can see them here, http://www.seventeenorbust.com/secret every once in a while the n increase jumps to +12.X M
If the first-pass queue runs dry, the sky will fall. The magic of centrifugal acceleration will be broken and the Earth will plummet into the inferno of the sun. Our souls will be cast into the outer darkness for eternal torment.
Er, maybe not that bad, but definitely bad. Right now, "normal" users (everyone except secret, etc.) are eligible for "global," "first-pass" and "dropped-tests" queues only. There's nothing in "global" and not much in "dropped-tests," so the server would quickly run out of tests and start denying work requests. All clients would run idle.
But don't worry, it won't happen! We're gonna repopulate the queue tonight =)
Re: double checks: I don't think we want to put all power on that. Not now and probably not ever. I do think second-pass is quite a bit behind, but I'm not sure of this. You're right, it depends on the error rate, and we haven't gone through those calculations recently. Probably should. The eventual goal would be to have the server assign exactly the number of second-pass tests needed (to normal users) to keep the n thresholds in the right proportion. With the right queue setup and server logic the two can be kept in lockstep. An advanced server could even keep a running tally of the error rate and adjust the thresholds accordingly.
Dunno when this will happen. I think for now we need to focus on getting a better client out. I think user retention probably has more to do with the client than anything else.
Your solution would be optimal and actually avoid/point out problems like the 2.2 verion etc. In the time being, have you considered manually adding ?100? tests perday for just this purpose? (I could actually do this myself if you wish) I don't think it would hurt the project...
The optimal for this of course would be to add one test every 15 minutes...
Fair enough. Sounds like a good game plan. Thx for the reply and keep up the good work. :thumbs:Quote:
Originally posted by kugano
The eventual goal would be to have the server assign exactly the number of second-pass tests needed (to normal users) to keep the n thresholds in the right proportion. With the right queue setup and server logic the two can be kept in lockstep. An advanced server could even keep a running tally of the error rate and adjust the thresholds accordingly.
Dunno when this will happen. I think for now we need to focus on getting a better client out. I think user retention probably has more to do with the client than anything else.
This hasn't been done yet - chances increase my guess will come nearest. :jester:Quote:
Originally posted by kugano
But don't worry, it won't happen! We're gonna repopulate the queue tonight =)
Hehe... yes, I got a bit carried away with some other code last night. Don't worry, it'll get done.
Thanks for clearing up the dropped-tests queue.Quote:
Originally posted by kugano
Hehe... yes, I got a bit carried away with some other code last night. Don't worry, it'll get done.
A bit more than 2 days left - it seems like Frodo will be the winner with a direct hit! :thumbs:Code:first-pass 53621 7977908 9999967 176 +10986
It looks like Frodo just won. This is the stat line now:
So we've just passed 8 MillionCode:first-pass 53002 8000887 9999967 113 +12400
Well done Frodo42 :)
Well that was a lucky punch :jester: :thumbs:
I guess it comes from all this factoring ... I haven't gotten one single factor from P-1 submitted to late yet ;)
You're still holding your own in 3rd ;)Quote:
I guess it comes from all this factoring ... I haven't gotten one single factor from P-1 submitted to late yet
Yes!
Congratulations Frodo!
it doesn't take long for them to start coming in.
I just got this one today:
[Mon Jan 17 16:53:06 2005] got proth test from server (k=19249, n=8012642)