PDA

View Full Version : When are we going to chage k



nuutti
10-07-2002, 04:03 PM
to k=67607 ?


Yours,

Nuutti

jjjjL
10-12-2002, 02:25 PM
k=67607?

after we work on k=33661, which will soon be reserved for us.

work on k=33661 will begin after the old versions of the client, namely v092, 094, and 096 are phased out. i want the verification data to be more complete on the next test.

i'm posting something about it on the website now.


-L

smh
10-12-2002, 03:04 PM
Originally posted by jjjjL (on the website)
The server has been seeded with several weeks worth of values for k=27653. Once we reach 3 million, we will begin testing k=33661. I have already begun farming out the factoring work to my computing cluster at U of M. For those interested, the newest n values on the server have been factored to over 120 billion. I hope to factor the next batch higher, possibly 1 trillion.


Why K=33661 ?

67607 is already reserved for SB for a long time and is the least far tested of the remaining exponents.

Also 67607 is already trialfactored to 178 G

BTW, according to Phil Carmody's page http://fatphil.org/maths/sierpinski/ 27653 is already sieved to 258 G

Are these save files used??

jwolfe
10-12-2002, 07:37 PM
Why not test more than one value of k at once? I would rather see all of them increase a little than have one increase a lot while another didn't increase at all. Plus, when a prime is found, there will be less wasted effort (or dead time) before the next exponent(s) takes over.

jjjjL
10-12-2002, 11:39 PM
smh - actually, i've read part of that site before, but i didn't notice that he had those files up there. I checked and he had eliminated 75 more n-values than I had by factoring to 120 Billion. I'm glad you mentioned that site to me, i had no idea that resource existed.

33661 is a much better candidate for searching with the current test we're using. 67607 will likely be the last factor to be finished because the expected prime n value is much larger. "reserved for SB" would be better read "no one has ever worked on it." as a matter of fact, Dave and I never requested that k value. Wilfrid Keller reserved it for us and just updated his website without telling us... we found out it was ours the same way you guys did... by happening to see it on his site one day. :rolleyes: i am still trying to decide when we may pursue that value. for now, i'd say 67607 is a non-issue.

jwolfe - i'm not doing multiple k values because other individuals are involved in private efforts to finish them. i like to share. sharing is good. also, a lot of factoring work has to be done before assigning k values since the client no longer factors on its own. also, in some ways, the testing methodology will be better for the next k value we test. we will have full verification data, etc.


-Louie

smh
10-13-2002, 07:39 AM
Originally posted by jjjjL
smh - actually, i've read part of that site before, but i didn't notice that he had those files up there. I checked and he had eliminated 75 more n-values than I had by factoring to 120 Billion. I'm glad you mentioned that site to me, i had no idea that resource existed.

I thought Phil would at least have told the people who had the exponents reserved. Any way, it saves a few tests.


Originally posted by jjjjL
33661 is a much better candidate for searching with the current test we're using. 67607 will likely be the last factor to be finished because the expected prime n value is much larger.


How can you expect where the expected prime will be? If you could, wouldn't that make the whole sierpinski problem a lot easier?


Originally posted by jjjjL
"reserved for SB" would be better read "no one has ever worked on it." as a matter of fact, Dave and I never requested that k value. Wilfrid Keller reserved it for us and just updated his website without telling us... we found out it was ours the same way you guys did... by happening to see it on his site one day. :rolleyes: i am still trying to decide when we may pursue that value. for now, i'd say 67607 is a non-issue.


Then maybe you should let someone else take this number. I guess at the moment nobody is working on this number, and as long as Wilfrid's site says its reserved for SB noone will.

One more thing, which i forgot earlier on, How long do you think it will take to reach n=3M? Don't forget that larger exponents take a lot longer to test, especially when you have to switch to a larger FFT. Which i guess will be around 2,6M if you use Woltmans code

jjjjL
10-13-2002, 01:40 PM
there are 2445 more n values to check before 3 million is reached as of Oct 13th.

the network's current rate is about 300tests/week. that may decrease a little, but that assumes weak growth (or almost none) in the # of users. there may also be a slight hit from people not upgrading and being blocked when i make the 0.9.7 client mandatory.

i'd guess the current k will take us the rest of 2002. hopefully in this time, I have time to finish teams and a few other improvements so that i can get SB ready for prime-time... more users would certainly help. don't forget, there is always the possibility that we find the prime too. :D

-L

smh
10-14-2002, 12:10 PM
there are 2445 more n values to check before 3 million is reached as of Oct 13th.

Can you make info like this available on the stats page? Also, how many exponents below 2M. Currentely current tests are >2M but the lowest number that hasn't been tested is 1,55M. No idea what the lowest one will be after this one is completed (how long befor it expires? etc,etc...)

Firebirth
10-15-2002, 02:19 PM
Why not test more than one value of k at once? I would rather see all of them increase a little than have one increase a lot while another didn't increase at all. Plus, when a prime is found, there will be less wasted effort (or dead time) before the next exponent(s) takes over.

Further... I don't think that very many new users with home computers will start to participate, if they are going to wait a couple of days to see any progress (I'm testing my first n on the second day and have only crunched 25%! :bang: ).

Therefore it would be nice if users with slow computers could be assigned low n (and hence another k value), and let the big numbers be left for servers and patient pc's

smh
10-16-2002, 05:15 AM
I installed the latest client on my notebook (Celeron 400) which is running during office hours (when i'm in the office and not at a client). The program reports that it takes a bit over 100 hours to test the current number (2,07M). I'm afraid this notebook won't be running 100+ hours in two weeks. When N gets larger it will only get harder to finish within the current two weeks period. If the period won't be extended for lets say another week i won't be able to use this notebook anymore.

jjjjL
10-17-2002, 08:45 PM
smh, all - i have a way to solve it. will be done in a few weeks when i get out from underneith the pile of work i have for classes.

i think it will be very cool... we'll see.

-L

smh
10-18-2002, 03:24 AM
Okay, but for the time being, is there a way to transfer a test from one pc to another? Just copying the save file didn't do. I guess there are some registry settings, but i didn't have time to figure that out