| Author |
Message |
|
FloppusMaximus
Advanced Member
Joined: 22 Aug 2008
Posts: 472
|
 Posted: 24 Aug 2009 02:13:28 am Post subject: |
|
|
This is one of the three main reasons I've been saying all along that we need a validator. (The others being (a) the 64-bit experimental siever, and (b) the enormous amount of CPU time that we are currently wasting by duplicating results.)
The GGNFS siever does allow resuming, and we are using that feature in the BOINC client as well. But it works in a fairly simple way: it looks at the relations that have already been written to the output file, determines what q value the siever was working on when it was interrupted, and proceeds from there. It does not discard existing relations that have already been found for that q value, so if all goes well, you should see those relations duplicated.
If we were using a proper validator, this would work out just fine - there is nothing wrong with having a few duplicated relations. I suppose if you really wanted to, you could add yet another ugly hack that would try to discard these duplicates. But I think it would be much easier just to use a validator!
Last edited by Guest on 24 Aug 2009 02:18:37 am; edited 1 time in total |
|
| Back to top |
|
|
squalyl
Advanced Newbie
Joined: 04 Aug 2009
Posts: 57
|
| Posted: 24 Aug 2009 03:16:10 am Post subject: |
|
|
well, ok.
will do asap, but I've professionnal work before.
Last edited by Guest on 24 Aug 2009 03:24:38 am; edited 1 time in total |
|
| Back to top |
|
|
frmky
Newbie
Joined: 22 Aug 2009
Posts: 5
|
| Posted: 24 Aug 2009 12:17:24 pm Post subject: |
|
|
| Checkpointing is as simple as saving the last special_q done. Restarting is simply starting again at the saved special q. Don't worry about the edge cases. If one special q gets skipped, it's not a big deal. There are lots more. If a special q gets duplicated, it's not a big deal. The nature of lattice sieving produces lots of duplicate relations anyway, so a few more won't hurt. If somehow a few relations become corrupted, again it's not a big deal. msieve checks that all relations are valid before it uses them, and discards the bad ones. (This discarding of bad relations is the Error -# when reading relations. Although it looks scary, it's not a problem.) In this regard, lattice sieving is much more forgiving than other projects. If a user submits bad data, it does not hurt the rest of the computation. |
|
| Back to top |
|
|
CompWiz
Advanced Newbie
Joined: 16 Oct 2005
Posts: 66
|
| Posted: 24 Aug 2009 03:11:19 pm Post subject: |
|
|
I just set up the BOINK client on my computer for this. I've got a 2.26ghz Core2Duo laptop. Also, I'm using the 64 bit client, so if you've got a faster 64 bit algorithm, I can run it.
It seems that there are no compute tasks up at the moment. If you do need more compute power, I can also run this on an old AMD X2 64 4400+, an AMD quad @3ghz(64 bit XP), and an Intel Core2Duo @3ghz. And, I've got another quad-core computer coming soon, which will replace the old AMD X2.
I bet this would run much faster if you could run it on a GPU, wouldn't it? Any chance you could crack the nspire? |
|
| Back to top |
|
|
FloppusMaximus
Advanced Member
Joined: 22 Aug 2008
Posts: 472
|
| Posted: 24 Aug 2009 11:08:30 pm Post subject: |
|
|
We must be getting close on several of the keys now, judging by the size of the .tar files.
CompWiz: Thanks for your help. Perhaps your quad-core machine could help with final processing on one of the keys that's nearing completion?
If you want to write or port a siever to run on GPUs, be my guest. :)
The Nspire? No, no chance.
The best polynomial so far for the 02 key has a score of 2.51e-12. We can add that to the BOINC project if there's any danger of running out of work. 
Last edited by Guest on 24 Aug 2009 11:13:11 pm; edited 1 time in total |
|
| Back to top |
|
|
Lionel Debroux
Member
Joined: 01 Aug 2009
Posts: 170
|
| Posted: 25 Aug 2009 01:20:52 am Post subject: |
|
|
Yeah, the 89T and V200 keys will soon be sieved enough. We don't understand what's going on with some areas of the 89 and 92+ keys, whose yields ([number of relations] / [number of q values]) are about four times smaller than on other areas (~0.3 instead of ~1.3-1.4). We might have to re-crunch these areas.
If you post the best .poly for 02 found by your computer, squalyl can add it to the BOINC project in the next few days 
On yAronet, I have already posted the polynomials for 0101 (from polys.tar.gz, selection ran up to ~20.8M leading coefficients), 0103 and 0109 (from my polynomial selection runs, up to ~25M leading coefficients for both). |
|
| Back to top |
|
|
FloppusMaximus
Advanced Member
Joined: 22 Aug 2008
Posts: 472
|
| Posted: 25 Aug 2009 02:03:55 am Post subject: |
|
|
[s]Best polynomial so far (I've searched coefficients up to ~22 million):[/s] (I've found a better one, see below)
Code:
n: 12778113034099956301555350695932258878071536472745082789663875691581271358898909
226606302252511075939436198764622302618337130658446299087762727153678760321
c5: 19150740
c4: -71666780933805
c3: 3320405629580867668
c2: 2759510053385868736054490
c1: 114465220521308121298479377238
c0: 2905124268495886593179284033445925
Y1: 553453762626682621
Y0: -231663177924282665664677692016
skew: 234527.60
lpbr: 29
lpba: 29
rlim: 27000000
alim: 27000000
mfbr: 58
mfba: 58
rlambda: 2.6
alambda: 2.6
(rlim/alim are just a guess)
Last edited by Guest on 25 Aug 2009 03:48:11 pm; edited 1 time in total |
|
| Back to top |
|
|
Nvidiot
Newbie
Joined: 25 Aug 2009
Posts: 2
|
| Posted: 25 Aug 2009 05:28:07 am Post subject: |
|
|
| I've joined BOINC with an i7 920. I am on Windows 7 64-bit, but the program that is doing the cracking is a 32-bit version. Is there a 64-bit version available and if so, how do I use it? |
|
| Back to top |
|
|
Lionel Debroux
Member
Joined: 01 Aug 2009
Posts: 170
|
| Posted: 25 Aug 2009 05:47:52 am Post subject: |
|
|
Nice computer
No 64-bit version available in the BOINC project yet. |
|
| Back to top |
|
|
CompWiz
Advanced Newbie
Joined: 16 Oct 2005
Posts: 66
|
| Posted: 25 Aug 2009 08:48:51 am Post subject: |
|
|
FloppusMaximus wrote: CompWiz: Thanks for your help. Perhaps your quad-core machine could help with final processing on one of the keys that's nearing completion?
Sure. Just tell me what I need to do.
FloppusMaximus wrote: If you want to write or port a siever to run on GPUs, be my guest.
I wish. I've started looking into GPU programming, but I haven't had much time to devote to it. |
|
| Back to top |
|
|
squalyl
Advanced Newbie
Joined: 04 Aug 2009
Posts: 57
|
|
| Back to top |
|
|
Nvidiot
Newbie
Joined: 25 Aug 2009
Posts: 2
|
| Posted: 25 Aug 2009 02:57:38 pm Post subject: |
|
|
Hmm, the processed work units are uploaded and then stay on the drive. Is that supposed to happen? In a few hours it has eaten up 5.5 MB, I'd rather not have this grow and grow and grow and grow :)
Anything you can do on the boinc-side of things or do I have to go somewhere on my drive and zap some files manually? |
|
| Back to top |
|
|
FloppusMaximus
Advanced Member
Joined: 22 Aug 2008
Posts: 472
|
| Posted: 25 Aug 2009 03:47:13 pm Post subject: |
|
|
No idea.
I've just found several better polynomials for the 02 key! The best one is
Code:
skew: 563358.35
c5: 23033160
c4: 290688239808
c3: -33765344004657556632
c2: -895543212198728821429709
c1: 3366563206170392319415115600982
c0: 164419144439246310238531580811313319
y1: 407059878063840287
y0: -223266023920651861709450622132
|
|
| Back to top |
|
|
CompWiz
Advanced Newbie
Joined: 16 Oct 2005
Posts: 66
|
| Posted: 25 Aug 2009 05:53:22 pm Post subject: |
|
|
squalyl wrote: with cuda or <insert ati framework name here> 
OpenCL works on nVidia and ATI cards, as well as normal CPU's. Why use a proprietary language?
I got the AMD quad @ 3ghz and the older AMD X2 working on the BOINK project. As I said, let me know if you'd like me to run something else.
Last edited by Guest on 25 Aug 2009 08:22:50 pm; edited 1 time in total |
|
| Back to top |
|
|
dga
Newbie
Joined: 20 Aug 2009
Posts: 7
|
| Posted: 25 Aug 2009 10:40:37 pm Post subject: |
|
|
FloppusMaximus wrote: If you want to write or port a siever to run on GPUs, be my guest. :)
The Nspire? No, no chance.
My understanding of the nspire situation is that people haven't figured what encryption or digital signatures are being used for the nspire code. I didn't follow enough of the threads back to notice why people suspect it's a 64-bit encryption algorithm, but -- if that's the case, it might not actually be that hard to find the symmetric key through a BOINC-like approach. If distributed.net could do it in 2002, surely seven years of Moore's law should make the problem a wee bit more tractable. It took them about 1700 days, and their computing power was only about the equivalent of a thousand dual-cpu, quad-core modern machines. But it sounds like there's a lot of work to be done -before- the brute force people get involved. |
|
| Back to top |
|
|
squalyl
Advanced Newbie
Joined: 04 Aug 2009
Posts: 57
|
| Posted: 26 Aug 2009 02:10:58 am Post subject: |
|
|
Nvidiot: sometimes workunit are uploaded but not reported. Updating the project fixes things. (5.5MB is ~3-4 workunits )
I don't know why this happens, the behaviour is not the same on all computers.
dga: 64 bits, ok, but what attack will you run? do we know the alignment? headers before encrypted data? do we have a known plain text ? initialization vectors? do they use derived symmetric keys using a weird scheme and a serial number? etc... |
|
| Back to top |
|
|
Lionel Debroux
Member
Joined: 01 Aug 2009
Posts: 170
|
| Posted: 26 Aug 2009 04:17:04 am Post subject: |
|
|
Has the 0104 key (mentioned at page 2, http://www.unitedti.org/index.php?showtopi...st&p=135126 ) already been factored somehow, or is it among the ones that we could add for the next batch of keys of the BOINC project, along with 0103, 0101 and 02 ?
Also, while the BOINC project is hot, do you think we should we use the opportunity to factor 0108, 0109 and 010A, even if we don't _need_ those keys because 0101, 0103 and 0104 do the job ?
On the one hand, I think it's interesting to factor _all_ interesting keys while we have a dedicated infrastructure for that task; on the other hand, if we can do without those three keys, we'd be factoring them just for fun. |
|
| Back to top |
|
|
yoyo
Newbie
Joined: 24 Aug 2009
Posts: 3
|
| Posted: 26 Aug 2009 10:16:00 am Post subject: |
|
|
Hello,
I would be interested if your infrastructure is able to factor any number and what are the constraints?
yoyo |
|
| Back to top |
|
|
squalyl
Advanced Newbie
Joined: 04 Aug 2009
Posts: 57
|
| Posted: 26 Aug 2009 11:26:22 am Post subject: |
|
|
Well, I can't answer that. We totally rely on the GNFS/Msieve projects, you may find information here, or better, wait FloppusMaximus explanations.
I think, given we broke a 512 bits key, that we can factor any number in the 1-512 bits range
maybe more bits are possible, but may require an increase in time / needed power.
the v200 has now 57 719 857 relations and the titanium has 54 409 443.
Who has the bigger CPUs? |
|
| Back to top |
|
|
yoyo
Newbie
Joined: 24 Aug 2009
Posts: 3
|
| Posted: 26 Aug 2009 11:47:18 am Post subject: |
|
|
It would mean, that you are able to factor e.g. this 149 digit number?
Quote: 23373956054212729369314132440470881224763236407256033079439078784258944564500359
718028342226892427079860170972262736474225014034561266391630623618447
Can we test it?
yoyo |
|
| Back to top |
|
|
|
Open Topic & United-TI Talk => General Open Topic
