Error reading relation

jux · 2016-02-06, 04:28

I am currently on the -nc step, and YAFU currently spits out some lines like error -15 reading relation 32507626, with all other millions of relations I presume being read correctly. Is this normal?

I cannot run -nc2 without getting this error: read_cycles can't open cycle file. To my knowledge, I have completed -np, -ns, and -nc.

LaurV · 2016-02-06, 05:17

Occasional relations can be bad, this is caused by a race condition when relations are written in the file, numbers of one relation can be missed or interleaved with numbers from other relation. If more than (guessing) 1% of the relations give you reading error, then you may have a problem (like it happened to me in the past I was sieving two different polys) but if only few (few hundreds, few thousand) relations are bad, it is of no concern.

edit: if you are really nitpicking about, and have a text editor that support such large files, then just identify the bad relations, go into the file and erase the lines

Also, forcing a filtering step will eliminate your duplicate relations, and the bad one will go lost together with them.

jux · 2016-02-06, 05:37

Alright, I'm pretty sure the issue with the cycle file does not have to do with reading relations. Do you know how you usually get a cycle file?

Dubslow · 2016-02-06, 07:30

It should be automatically created as part of the -nc process (I'm not certain which part). I would say it's probably an issue with file permissions, though that should have make itself apparent long before you got to -nc.

jux · 2016-02-06, 07:48

It seems like all the steps, including -nc, finished correctly (end of -nc in screenshot), so I'm not sure what the problem is here. I guess I'll give my computer a break tonight.

LaurV · 2016-02-06, 08:08

Quote:

Originally Posted by jux

It seems like all the steps, including -nc, finished correctly (end of -nc in screenshot), so I'm not sure what the problem is here. I guess I'll give my computer a break tonight.

You have much more ideals than relations. Are you over/under-sieved? You were ok before the last step, it seems. Can you cut out about 5-10% of the relations file and run again, with resume? It may possible try to sieve more, etc. But it is a pity to throw away the work already done.

Dubslow · 2016-02-06, 08:31

Quote:

Originally Posted by jux

It seems like all the steps, including -nc, finished correctly (end of -nc in screenshot), so I'm not sure what the problem is here. I guess I'll give my computer a break tonight.

-nc did not finish correctly, in fact you didn't even get to -nc2.

jux · 2016-02-07, 08:57

Quote:

Originally Posted by LaurV

You have much more ideals than relations. Are you over/under-sieved? You were ok before the last step, it seems. Can you cut out about 5-10% of the relations file and run again, with resume? It may possible try to sieve more, etc. But it is a pity to throw away the work already done.

I don't know about over/under-sieved, but -ns seemed to run to completion and the nfs.dat file is 5.8 GB, which is about the size of the .dat file I have from a successful C150 factoring. I don't even know how to cut 5-10% from the relations file of that size - what program do I even use? I guess file.truncate with python (or sed, or some windows batch trick).
I think this may be just an issue with YAFU - my C150 attempt used factmsieve.py and worked fine.

Dubslow · 2016-02-07, 09:34

38M unique relations sounds kind of low for a C155. Can you post the contents of your nfs.job file, and try continuing with -ns instead of -nc? I don't think you're done sieving.

If LaurV is right, then there is a truncate command on most Linuxes, or the Python you mentioned would work too.

jux · 2016-02-07, 09:46

Here is nfs.job, which I remember manually editing . That may have caused problems.
Anyway I figured out how to truncate by lines using wc and sed. So I'll try sieving more.

Code:

n: 10941738641570527421809707322040357612003732945449205990913842131476349984288934784717997257891267332497625752899781833797076537244027146743531593354333897
skew: 71994382.30
Y0: -3137301482625007833545663308178
Y1: 58950047158989383
c0: 61702731036896719906612929448642480018731
c1: -4099166098410202538646695681603748
c2: -32522921176495846591845785
c3: 2916461378816208604
c4: 11304125346
c5: 36

rlim: 32000000
alim: 32000000
lpbr: 29
lpba: 29
mfbr: 58
mfba: 58
rlambda: 2.6
alambda: 2.6

debrouxl · 2016-02-07, 10:37

The parameters look alright, but 38M unique relations is indeed too low for a C155 (e.g. a 512-bit semi-prime). 5M unique relations is about the bare minimum for building a matrix, and 5M more unique relations doesn't hurt.

2016-02-06, 04:28	#1
jux Aug 2015 36₁₆ Posts	Error reading relation I am currently on the -nc step, and YAFU currently spits out some lines like error -15 reading relation 32507626, with all other millions of relations I presume being read correctly. Is this normal? I cannot run -nc2 without getting this error: read_cycles can't open cycle file. To my knowledge, I have completed -np, -ns, and -nc. Last fiddled with by jux on 2016-02-06 at 04:56

2016-02-06, 05:17	#2
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 19×541 Posts	Occasional relations can be bad, this is caused by a race condition when relations are written in the file, numbers of one relation can be missed or interleaved with numbers from other relation. If more than (guessing) 1% of the relations give you reading error, then you may have a problem (like it happened to me in the past I was sieving two different polys) but if only few (few hundreds, few thousand) relations are bad, it is of no concern. edit: if you are really nitpicking about, and have a text editor that support such large files, then just identify the bad relations, go into the file and erase the lines Also, forcing a filtering step will eliminate your duplicate relations, and the bad one will go lost together with them. Last fiddled with by LaurV on 2016-02-06 at 05:24

2016-02-06, 07:48	#5
jux Aug 2015 66₈ Posts	It seems like all the steps, including -nc, finished correctly (end of -nc in screenshot), so I'm not sure what the problem is here. I guess I'll give my computer a break tonight. Attached Thumbnails Last fiddled with by jux on 2016-02-06 at 07:49

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
error: cannot locate relation	cardmaker	Factoring	16	2017-07-17 12:38
msieve C157 sqrt error: relation xxxx corrupt	tgrdy	Msieve	6	2010-08-20 21:51
Error reading intermediate file	sharad	Software	0	2004-08-09 10:40
Error reading intermediate file	edorajh	Software	22	2004-01-29 22:13
ERROR: reading from temp file	T-bird	Hardware	1	2002-12-05 16:35

2016-02-06, 05:37	#3
jux Aug 2015 2·3³ Posts	Alright, I'm pretty sure the issue with the cycle file does not have to do with reading relations. Do you know how you usually get a cycle file?

2016-02-06, 07:30	#4
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3×29×83 Posts	It should be automatically created as part of the -nc process (I'm not certain which part). I would say it's probably an issue with file permissions, though that should have make itself apparent long before you got to -nc.

2016-02-07, 09:34	#9
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 1110000110101₂ Posts	38M unique relations sounds kind of low for a C155. Can you post the contents of your nfs.job file, and try continuing with -ns instead of -nc? I don't think you're done sieving. If LaurV is right, then there is a truncate command on most Linuxes, or the Python you mentioned would work too.

2016-02-07, 10:37	#11
debrouxl Sep 2009 11·89 Posts	The parameters look alright, but 38M unique relations is indeed too low for a C155 (e.g. a 512-bit semi-prime). 5M unique relations is about the bare minimum for building a matrix, and 5M more unique relations doesn't hurt.