AFAIR we had this bug once before. It did turn out to be only the last digit of the found factor, didn't it?
Well I thought about that possibility. I'll let the client run through the test for that number from the beginning again when I return to the computer.AFAIR we had this bug once before. It did turn out to be only the last digit of the found factor, didn't it?
I can't make the P-1-factorer give any output but 7 | 5359*2^4754206+1 for this test. Probably only the last digit, but how do I get my hands on the rest of the digits
You might have a prime!Originally posted by Frodo42
I can't make the P-1-factorer give any output but 7 | 5359*2^4754206+1 for this test. Probably only the last digit, but how do I get my hands on the rest of the digits
I would suggest running PRP or Primo on it just to be sure.
Joe O
is there a particular reason why that would indicate a prime?
I've been debugging programs for nearly 41 years. This is the type of "bug" that I've seen many a time. "We are only looking at composite numbers to factor, we'll never not have 2 factors." Or it could just be a very large factor, larger than the program expects to discover.
If it is indeed the last number of the "factor", the fact that it is a 7 certainly does not rule out the possibility that it is a prime. If you recall, wblipp has shown that all of the remaining primes will end in 7.
Joe O
Frodo42,
Did you try rediscovering your factor with different bounds?
What would happen if you used bounds that are too small to find the unknown factor?
What happens, if you continue using the safefile of one of those tests?
Nevertheless I wish you good luck
I tried with lots of difefrent bounds and redoing it different ways.Did you try rediscovering your factor with different bounds?
What would happen if you used bounds that are too small to find the unknown factor?
What happens, if you continue using the safefile of one of those tests?
With weight smaller than 1.4 (can't remember the specific bounds) it just does not find anything and goes on to stage 2 where it also doesn't find anything.
I also tried with lower bounds and then raising them on the next run, that still gives me that blasted 7 ...
Anyone who can find the right factor is welcome to do so and take the credit in the stats. I've given up and use my processor power for continued P-1 factoring instead.
Well we know it is not prime.
Code:5359*2^4754206+1 has no small factor. 5359*2^4754206+1 is composite: [2CC06F27C7AF165D] (695868.8883s+0.0115s)
Joe O
I want to run the client on my P4, but unfortunately I can not.
I don't know why, but the client exits when I try to factor a range. To be more precise, it either exits (for ranges above 4983000, if a I recall correctly), or gives output like
for ranges below 4983000 (I also doubt if this output is ok. As I recall, it took longer, and gave % completion info as well.)Expected number of factors for entire range: 0.089770
B1=20000 B2=200000 Success=0.008977 Squarings=45863
P-1 on 33661*2^4495032+1 with B1=20000, B2=200000
initializing test
sieve finished
33661*2^4495032+1 stage 1 complete. 188 transforms. Time: 21 seconds
Starting stage 1 GCD - please be patient.
33661*2^4495032+1 stage 2 complete. 226 transforms. Time: 301 seconds
Starting stage 2 GCD - please be patient.
finished.
Total factoring Time: 10 minutes
PS:
I've tried various versions of the client,
I've read this thread a couple of times to find out what I'm doing wrong,
I've was able to run it in the beginning (4080000 4081000 Nuri, client version was 0.6 I guess).
I guess this is because I am skiping some step somewhere, but I can not recognise where.
Any comments is welcome.
The output looks good until we reach "sieve finished".
But then, the # of transforms should be in the 10,000s.
The calling syntax is
sbfactor.exe <start> <end> <whatWasThisFor?> <boundChanger> <allowedMemoryConsumption>
But I suspect yours is just fine...
Here's an example from one of the alternative ranges I tried. (This was to test if it finds the factor frodo42 found).
sbfactor.exe 4929741 4929743 46 1.5 128
46 is the sieve dept - 2^46 - (we've finished sieving of almost everything below 70T, so I tried 46 instead of 45)
or another one; sbfactor.exe 4980000 4982000 45 2 192
Both gives similar type of output (i.e. # transforms < 1,000)
Which version of the client are you using?
For most of the factors i found I used 45 sieve depth and 196 memory.sbfactor.exe 4929741 4929743 46 1.5 128
Anyways I run this client on Linux, but it also works fine with win98.
I use client version 1.1
I have not yet gone to ranges above 4983000.
Sadly, I don't see what is going wrong here.
Is there probably a difference between 2 and 2.0?
That's strange. I've downloaded the client again, and it seems to work now (for n below 4981000. above that, it still exits). Unlike previous efforts, it now shows percent completion info. I am reserving a range close to that (4980000 4982000) to see what will happen.
Thanks for the ideas.
PS: I hope the exit problem for ranges above 4981000 is something related to what I might be doing wrong. Otherwise, it might be due to a bug in the program. While browsing through the pages, I noticed that ceselb also reported such a problem.
Originally posted by ceselb
Btw, anyone tried to do a range over 5M? Sbfactor just exits if I try
where does the progra moutput factors when they are found?
It writes them into a separate text file. I don't konw the exact file name ATM, as I haven't found any factors the last couple of days.
But I'm sure you'll find the file once factors are found.
edit:
Originally posted by Joe O
fact.txt
Oh, and I also have the same problem (crash) when trying to factor sth. bigger than 5359*2^4980726+1...
Last edited by Mystwalker; 10-15-2003 at 08:39 PM.
.
Joe O
The executable dated 7/2/03 4:35am seems to work for ranges above 5000000. At least on a PIII/500 under Win98SE.Originally posted by Nuri
That's strange. I've downloaded the client again, and it seems to work now (for n below 4981000. above that, it still exits). Unlike previous efforts, it now shows percent completion info. I am reserving a range close to that (4980000 4982000) to see what will happen.
Thanks for the ideas.
PS: I hope the exit problem for ranges above 4981000 is something related to what I might be doing wrong. Otherwise, it might be due to a bug in the program. While browsing through the pages, I noticed that ceselb also reported such a problem.
Joe O
Ijust started running the client today and it finished the first part of the first test in 3791 seconds but the seond part is taking alot longer Iunderstand this is right but how long should the second part take. It has only reported about 2 % progress adn its ben running for a long while now. Ihave an AMD2000+ with 256 MB DDR. The resource manager says I'm using very little of my CPU and alot of memory. Is this supposed ot take this long.
The short answer is "yes".
The long answer is:
What is your command line? especially the last parameter.
e.g.
sbfactor.exe 4929741 4929743 46 1.5 128
The last parameter 128 says to use 128MBytes of ram for the second stage of the P-1. You *might* be able to specifiy 128 on your 256MB machine, as long as you are not running anything else. Anything higher and you will *thrash*, i.e. page back and forth from memory and the swap file on your hard drive without doing any useful work.
I almost forgot to ask: Do you have a separate video card, or is the video built in? If built in, how much "shared" memory is it using?
Joe O
I think i was running 256 I'll try changin it
This is the same executable that I downloaded on my last trial. It's probably either not working on PIVs, or the problem is specific to my box.Originally posted by Joe O
The executable dated 7/2/03 4:35am seems to work for ranges above 5000000. At least on a PIII/500 under Win98SE.
I'll try it on AMD Athlon and post the case for it.
This is the same executable that I downloaded on my last trial. It's probably either not working on PIVs, or the problem is specific to my box.Originally posted by Joe O
The executable dated 7/2/03 4:35am seems to work for ranges above 5000000. At least on a PIII/500 under Win98SE.
I'll try it on AMD Athlon and post the case for it.
Seems like it's a problem with P4s. I have the mentioned version of sbfactor, too. And it does work on my P3...Originally posted by Nuri
This is the same executable that I downloaded on my last trial. It's probably either not working on PIVs, or the problem is specific to my box.
Yes, it seems so. I've just tried on Athlon 2400, and it works fine.
As far as I remember, it was mentioned somewhere that PIV is the most efficient machine for this task. Too bad if it actually is the case.
I hope Louie can have the time to take a look at it.
Nuri, Have you had a chance to try out the two "beta" versions? They are descirbed in Louie's July 14th post in this thread. I would be very interested to know if either or both of them worked.
Just above that is the following from Mystwalker:
"btw. version 1.1 did work on the P4 system that crashed with the old version."
Is that still true? Mystwalker?
Joe O
Unfortunately, I don't have access to that system anymore. But I guess it's the same as other P4 systems: Up to almost 5M, it works.Originally posted by Joe O
Just above that is the following from Mystwalker:
"btw. version 1.1 did work on the P4 system that crashed with the old version."
Is that still true? Mystwalker?
I'll try the developer version tomorrow. Maybe that one can cross this frontier...
does anyone know thename of the file that factors are written to?
The factors are written in cut-'n-paste format to:
fact.txt
it writes to the end of the file, so if your siever has left a fact.txt in the same directory, the results will be on the end of the file.
when it finds a factor is it at the end of the process of factoring or does it stop in the middle say at 56%. Just wondering.
It reports factors at the end of the GCD process (at the end of the stage1 and stage2 tests).
I have been wondering if it would pay of to make GCD in the middle of the test's.
But I don't think that it would pay off, since it's pretty few factors that are found and it takes precious time from factoring.
The answer is still the same as it was when you asked this 2 days ago. "where does the progra moutput factors when they are found?"Originally posted by Keroberts1
does anyone know thename of the file that factors are written to?
FACT.TXT
Joe O
Keroberts, I guess you asked that question because you are running P-1 client on a different folder, and could not find the fact.txt file. Just a wild guess.
The client creates the fact.txt file right after it finds it's first factor. If it has not found any factors since you started, there will not be a fact.txt yet.
I hope this helps.
My main question ws wether or not it would write to the same file as proth sieve. Frod opointed out that it wil write to the same file and I think I'm good with that. I just need to remember to keep checking for any new factors at the end of that file.
That was one of the versions that I could not make work (1.1dev, and also 1.0, 0.9, 0.8, and 0.6). I did not notice previously that there was another 1.1 apart from 1.1dev. So, when I noticed that it was something different, I downloaded 1.1 and it simply worked (below 4.9 million etc).Originally posted by Joe O
Nuri, Have you had a chance to try out the two "beta" versions? They are descirbed in Louie's July 14th post in this thread. I would be very interested to know if either or both of them worked.
Having read the post you mentioned a couple of times more, I now understand that the client that I should have tried is ecm.exe, not 1.1dev. "If" I understood the post correctly, 1.1.dev is for finding larger factors of the k/n pairs that we know has some smaller factors (which we should write in the badsieve.txt file for the client to skip those). Although it is cool, that's not something I am particularly trying to do right now, so I'll skip 1.1dev. (PS: I might be wrong on this, but this is how I understood it.)
Anyways, I'll give ecm.exe a try and let you know the result.
Also, I'll try all these versions (1.1 and ecm) on another P4 - a notebook to see if it will work for larger numbers. If either of them works, that would be great. This machine is not connected to the net, and all I could do previously was to crunch the sieve client at 130kp/sec. It would perform definitely much better on P-1.
Nuri, I agree that ecm.exe is the one to try. 1.1dev without the badsieve.txt file should behave like 1.1. In fact it may be the same as 1.1. If ecm.exe doesn't work, then 1.1dev without the file may be worth a try, just in case it is different from 1.1. (Oops, i see that you did already try it to no avail!) This is looking more and more like an FFT boundary issue. Looking forward to hearing your results.
Joe O
Made the tests for two different P4 machines as posted above.
mostly bad news...
- The notebook was the third P4 machine that reports 1.1 exiting for n >= 4980726.
- Strangely, 1.1dev without badsieve does not behave like 1.1. Tried again, behaves like I explained on my post above. (total factoring time < 3 min, # transforms < 1,000, and not surprisingly, can not find the factors that 1.1 can find with the same factor value, factor depth, and memory settings), and exits for n>=4980726
- I'm not sure if I could run ecm correctly. I tried two alternatives.
ecm.exe <n low> <n high> <factor depth> <factor value> (mem usage at local.ini) ==> exits starting at 4980726.
ecm.exe k n B1 B2 #curves ==> seems to work for all n. (but, one can not enter a range. also, I'm not sure what values might be reasonable for B1, B2, and #curves. I used 120, 500 and 100, respectively. I dunno, may be I understood the usage wrongly. Any help is welcome).
I have done similar tests, with the same results. I've tried bigger n values to see if this is an FFT boundary issue, I tried 0.5M steps upto 7M, and all fail in the same way.Made the tests for two different P4 machines as posted above.
Is there anyone out there that has P-1 working for n>5M? If not, then it would appear that for a short while at least, we have lost a very valuable tool.
Originally posted by MikeH
Is there anyone out there that has P-1 working for n>5M? If not, then it would appear that for a short while at least, we have lost a very valuable tool.Mike, I think that the problem is only with PIVs. I have a PIII/500 merrily chugging away at 5007000 to 5008000. Joe.Originally posted by Joe O
The executable dated 7/2/03 4:35am seems to work for ranges above 5000000. At least on a PIII/500 under Win98SE.
Joe O
Posting Permissions
Reply With Quote