Preparing N=333333

[biwema]

Hi,

The chance that N=333333 will be our next exponent for twinprime search is quite good.
I did already quite some presieving and I think now it is the time to release the sievefile, that people who are interested, can play with.

At the moment the sieving limt is 205T; the range of the candidates is 100G.
There are 38335243 candidates left, but the removal rate is still more than 1 candidate per second (P4 3.4GHz)
The intermediate file is 727MB (762446156 Bytes). The zip file at maximum compression is 136MB (143112441 Bytes):

You can get it here:
http://www.transferbigfiles.com/Down...e-094891fc8cee

I have no own webspace available, therefore I use this service. The file can be downloaded 20 times in the next 5 days. Maybe someone of you has enough space to put it online.

I think that size is reasonable for the moment. In case of a large-scale distributed sieving one could write a dedicated software for the compression (the value 333333 only needs to be stored once; take only the gaps as multiple of 6 instead of the whole numbers, and zip the whole resulting file)

While sieving it could make sense just to synchronize the removed candidates. Note: You must not forget to turn on this feature in NewPGen every time you start sieving, otherwise comparing the intermediate files is necessary. Renaming the resulting file might prevent confusion (The exponent is not listed in the file)

The Expected number of remaining candidates is at...

1P 34.9 million
5P 31.8 million
10P 30.7 million
15P 30.0 million
20P 29.5 million
30P 28.9 million
50P 28.2 million

from 20P to 30P, about 6-7 years of sieving is necessary. 600000 candidtes are removed. LLR that many numbers needs about the same time (depending on the architecture). The optimal sieving limit will be in that area.

If someone wants to continue sieving (or already start with llr), we can synchronize our effort in this thread.

[smh]

I do have some hardware that can't run GNFS or the boinc client on which i can do some sieving, but some coordination is needed.

Are you continuing to sieve?
Will you collect all factors or are you looking for someone else to do that?

I'll post a range after i have a good idea on how long it will take to sieve.

[ValerieVonck]

Dear,

Oops too fast again.

You should have a look at Bittorrent for distributing the file.
My $0.02

Regards
Cedric.

[Rytis]

I have a torrent tracker running on PrimeGrid, it could be used. Contact me and I'll give you the instructions on how to prepare the .torrent file.

[biwema]

At the moment I stopped sieving. When I continue sieving, I will 'reserve' the range here on the forum to prevent duplicate sieving.

Bittorrent is one way to distribute it. I am not permanently online, so I am not the optimal person to start a torrent. Maybe someone else is more experienced in setting up a torrent.

[MooooMoo]

For sieving, how much RAM should people have? I'm considering participating in the sieving, but I only have 512 MB of RAM. And the uncompressed data file is > 700 MB

[biwema]

NewPGen needs between 300000 and 320000 K. (In the savefile one candidate uses 20 Byte or more, in the memory it is less)
I recommend at least 768M or better 1G. Probably 512M is also possible, but makes only sense if nobody is working on the computer while sieving. I did not try with 512M.

[ValerieVonck]

Shouldn't it be better that you split the monster file into multiple smaller one's?

E.g. dat file (or something else) that contain 10k -100K candidates and then a user can reserve that file for a specific sieve range from 0 till 1T or up??

[biwema]

Originally Posted by CedricVonck

Shouldn't it be better that you split the monster file into multiple smaller one's?

E.g. dat file (or something else) that contain 10k -100K candidates and then a user can reserve that file for a specific sieve range from 0 till 1T or up??

No.
The speed of horizontal sieving (fixed-k) does not depend on the size.

It does not take much longer (maybe a few %) to sieve a 100G than a 10k range from for example 1T to 2T. Just the saving of the intermediate file takes more time (because it is larger).

That means, that sieving a 10k range is 10000 times less efficient than sieving the whole 100G range.

That's why the presieving provides such a great speed-up to the twinprime search.

[pacionet]

I can offer myself as candidate to upload on the site (twinprimesearch.org) the 136M file for n=333,333 but only after January 3 (I'll be out some days for holidays ! ) and to continue the sieving.

So , when I return, I'll begin to prepare the new site with some n to sieve, even if I must check the disk space that they give me.

I Think that each file must be a SINGLE file. (one user sieving one n).

PLease tell me the exponents (after n=333,333) you want to use in our search.

Happy new Year to everybody !

Biwema, you sieved n=333,333 until which K : 0 to 25G ?

[MooooMoo]

pacionet,

I Think that each file must be a SINGLE file. (one user sieving one n).

PLease tell me the exponents (after n=333,333) you want to use in our search.

Various versions of this question has come up before, and the answer remains the same. I will not create any polls or threads discussing what n to sieve after n=333,333, until either:

1.) n=333,333 is already sieved to the optimal limit
2.) There is a twin found for n=195,000
3.) All LLR tests for n=195,000 have been completed to 25G, which means LLR testing is moving to n=333,333.

The problem with discussing the next few n's is that there are too many uncertainties, since it's too far in the future. For example, if any of these change:

- Number of participants
- Chip makers failing/exceeding Moore's law expectations
- Dedication of participants (people staying longer and/or putting more CPUs on the project)

then our premature vote for the next few n's means that those n's could either be set too high or too low.

Biwema, you sieved n=333,333 until which K : 0 to 25G ?

The range of k is 1-100G. The chance of finding a twin in that range is slightly greater than 90% (the chance of finding a twin in range 1-25G, with n=195000, is slightly under 90%).

---------------------

I'll be creating a new thread (sieve reservations thread for n=333,333) soon. It isn't likely that there will be more than 20 participants, so I'll be able to handle the load of updating the front page on a regular basis.

edit: There needs to be some kind of system (maybe a factor verification script?) to deter cheaters. This is quite important, since part of the credit for finding the twin will go to the top siever. Suggestions on preventing/deterring cheating are welcome.